2 * Copyright(c) 2017 Intel Corporation. All rights reserved.
4 * This program is free software; you can redistribute it and/or modify
5 * it under the terms of version 2 of the GNU General Public License as
6 * published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful, but
9 * WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * This code is based in part on work published here:
15 * https://github.com/IAIK/KAISER
17 * The original work was written by and and signed off by for the Linux
20 * Signed-off-by: Richard Fellner <richard.fellner@student.tugraz.at>
21 * Signed-off-by: Moritz Lipp <moritz.lipp@iaik.tugraz.at>
22 * Signed-off-by: Daniel Gruss <daniel.gruss@iaik.tugraz.at>
23 * Signed-off-by: Michael Schwarz <michael.schwarz@iaik.tugraz.at>
25 * Major changes to the original code by: Dave Hansen <dave.hansen@intel.com>
26 * Mostly rewritten by Thomas Gleixner <tglx@linutronix.de> and
27 * Andy Lutomirsky <luto@amacapital.net>
29 #include <linux/kernel.h>
30 #include <linux/errno.h>
31 #include <linux/string.h>
32 #include <linux/types.h>
33 #include <linux/bug.h>
34 #include <linux/init.h>
35 #include <linux/spinlock.h>
37 #include <linux/uaccess.h>
39 #include <asm/cpufeature.h>
40 #include <asm/hypervisor.h>
41 #include <asm/cmdline.h>
43 #include <asm/pgtable.h>
44 #include <asm/pgalloc.h>
45 #include <asm/tlbflush.h>
49 #define pr_fmt(fmt) "Kernel/User page tables isolation: " fmt
51 /* Backporting helper */
53 #define __GFP_NOTRACK 0
56 static void __init pti_print_if_insecure(const char *reason)
58 if (boot_cpu_has_bug(X86_BUG_CPU_INSECURE))
59 pr_info("%s\n", reason);
62 static void __init pti_print_if_secure(const char *reason)
64 if (!boot_cpu_has_bug(X86_BUG_CPU_INSECURE))
65 pr_info("%s\n", reason);
68 void __init pti_check_boottime_disable(void)
73 if (hypervisor_is_type(X86_HYPER_XEN_PV)) {
74 pti_print_if_insecure("disabled on XEN PV.");
78 ret = cmdline_find_option(boot_command_line, "pti", arg, sizeof(arg));
80 if (ret == 3 && !strncmp(arg, "off", 3)) {
81 pti_print_if_insecure("disabled on command line.");
84 if (ret == 2 && !strncmp(arg, "on", 2)) {
85 pti_print_if_secure("force enabled on command line.");
88 if (ret == 4 && !strncmp(arg, "auto", 4))
92 if (cmdline_find_option_bool(boot_command_line, "nopti")) {
93 pti_print_if_insecure("disabled on command line.");
98 if (!boot_cpu_has_bug(X86_BUG_CPU_INSECURE))
101 setup_force_cpu_cap(X86_FEATURE_PTI);
104 pgd_t __pti_set_user_pgd(pgd_t *pgdp, pgd_t pgd)
107 * Changes to the high (kernel) portion of the kernelmode page
108 * tables are not automatically propagated to the usermode tables.
110 * Users should keep in mind that, unlike the kernelmode tables,
111 * there is no vmalloc_fault equivalent for the usermode tables.
112 * Top-level entries added to init_mm's usermode pgd after boot
113 * will not be automatically propagated to other mms.
115 if (!pgdp_maps_userspace(pgdp))
119 * The user page tables get the full PGD, accessible from
122 kernel_to_user_pgdp(pgdp)->pgd = pgd.pgd;
125 * If this is normal user memory, make it NX in the kernel
126 * pagetables so that, if we somehow screw up and return to
127 * usermode with the kernel CR3 loaded, we'll get a page fault
128 * instead of allowing user code to execute with the wrong CR3.
130 * As exceptions, we don't set NX if:
131 * - _PAGE_USER is not set. This could be an executable
132 * EFI runtime mapping or something similar, and the kernel
133 * may execute from it
134 * - we don't have NX support
135 * - we're clearing the PGD (i.e. the new pgd is not present).
137 if ((pgd.pgd & (_PAGE_USER|_PAGE_PRESENT)) == (_PAGE_USER|_PAGE_PRESENT) &&
138 (__supported_pte_mask & _PAGE_NX))
141 /* return the copy of the PGD we want the kernel to use: */
146 * Walk the user copy of the page tables (optionally) trying to allocate
147 * page table pages on the way down.
149 * Returns a pointer to a P4D on success, or NULL on failure.
151 static p4d_t *pti_user_pagetable_walk_p4d(unsigned long address)
153 pgd_t *pgd = kernel_to_user_pgdp(pgd_offset_k(address));
154 gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
156 if (address < PAGE_OFFSET) {
157 WARN_ONCE(1, "attempt to walk user address\n");
161 if (pgd_none(*pgd)) {
162 unsigned long new_p4d_page = __get_free_page(gfp);
166 if (pgd_none(*pgd)) {
167 set_pgd(pgd, __pgd(_KERNPG_TABLE | __pa(new_p4d_page)));
171 free_page(new_p4d_page);
173 BUILD_BUG_ON(pgd_large(*pgd) != 0);
175 return p4d_offset(pgd, address);
179 * Walk the user copy of the page tables (optionally) trying to allocate
180 * page table pages on the way down.
182 * Returns a pointer to a PMD on success, or NULL on failure.
184 static pmd_t *pti_user_pagetable_walk_pmd(unsigned long address)
186 gfp_t gfp = (GFP_KERNEL | __GFP_NOTRACK | __GFP_ZERO);
187 p4d_t *p4d = pti_user_pagetable_walk_p4d(address);
190 BUILD_BUG_ON(p4d_large(*p4d) != 0);
191 if (p4d_none(*p4d)) {
192 unsigned long new_pud_page = __get_free_page(gfp);
196 if (p4d_none(*p4d)) {
197 set_p4d(p4d, __p4d(_KERNPG_TABLE | __pa(new_pud_page)));
201 free_page(new_pud_page);
204 pud = pud_offset(p4d, address);
205 /* The user page tables do not use large mappings: */
206 if (pud_large(*pud)) {
210 if (pud_none(*pud)) {
211 unsigned long new_pmd_page = __get_free_page(gfp);
215 if (pud_none(*pud)) {
216 set_pud(pud, __pud(_KERNPG_TABLE | __pa(new_pmd_page)));
220 free_page(new_pmd_page);
223 return pmd_offset(pud, address);
227 pti_clone_pmds(unsigned long start, unsigned long end, pmdval_t clear)
232 * Clone the populated PMDs which cover start to end. These PMD areas
235 for (addr = start; addr < end; addr += PMD_SIZE) {
236 pmd_t *pmd, *target_pmd;
241 pgd = pgd_offset_k(addr);
242 if (WARN_ON(pgd_none(*pgd)))
244 p4d = p4d_offset(pgd, addr);
245 if (WARN_ON(p4d_none(*p4d)))
247 pud = pud_offset(p4d, addr);
250 pmd = pmd_offset(pud, addr);
254 target_pmd = pti_user_pagetable_walk_pmd(addr);
255 if (WARN_ON(!target_pmd))
259 * Copy the PMD. That is, the kernelmode and usermode
260 * tables will share the last-level page tables of this
263 *target_pmd = pmd_clear_flags(*pmd, clear);
268 * Clone a single p4d (i.e. a top-level entry on 4-level systems and a
269 * next-level entry on 5-level systems.
271 static void __init pti_clone_p4d(unsigned long addr)
273 p4d_t *kernel_p4d, *user_p4d;
276 user_p4d = pti_user_pagetable_walk_p4d(addr);
277 kernel_pgd = pgd_offset_k(addr);
278 kernel_p4d = p4d_offset(kernel_pgd, addr);
279 *user_p4d = *kernel_p4d;
283 * Clone the CPU_ENTRY_AREA into the user space visible page table.
285 static void __init pti_clone_user_shared(void)
287 pti_clone_p4d(CPU_ENTRY_AREA_BASE);
291 * Initialize kernel page table isolation
293 void __init pti_init(void)
295 if (!static_cpu_has(X86_FEATURE_PTI))
298 pr_info("enabled\n");
300 pti_clone_user_shared();