xref: /openbmc/linux/arch/x86/include/asm/pgtable_64.h (revision addee42a)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PGTABLE_64_H
3 #define _ASM_X86_PGTABLE_64_H
4 
5 #include <linux/const.h>
6 #include <asm/pgtable_64_types.h>
7 
8 #ifndef __ASSEMBLY__
9 
10 /*
11  * This file contains the functions and defines necessary to modify and use
12  * the x86-64 page table tree.
13  */
14 #include <asm/processor.h>
15 #include <linux/bitops.h>
16 #include <linux/threads.h>
17 
18 extern p4d_t level4_kernel_pgt[512];
19 extern p4d_t level4_ident_pgt[512];
20 extern pud_t level3_kernel_pgt[512];
21 extern pud_t level3_ident_pgt[512];
22 extern pmd_t level2_kernel_pgt[512];
23 extern pmd_t level2_fixmap_pgt[512];
24 extern pmd_t level2_ident_pgt[512];
25 extern pte_t level1_fixmap_pgt[512];
26 extern pgd_t init_top_pgt[];
27 
28 #define swapper_pg_dir init_top_pgt
29 
30 extern void paging_init(void);
31 static inline void sync_initial_page_table(void) { }
32 
33 #define pte_ERROR(e)					\
34 	pr_err("%s:%d: bad pte %p(%016lx)\n",		\
35 	       __FILE__, __LINE__, &(e), pte_val(e))
36 #define pmd_ERROR(e)					\
37 	pr_err("%s:%d: bad pmd %p(%016lx)\n",		\
38 	       __FILE__, __LINE__, &(e), pmd_val(e))
39 #define pud_ERROR(e)					\
40 	pr_err("%s:%d: bad pud %p(%016lx)\n",		\
41 	       __FILE__, __LINE__, &(e), pud_val(e))
42 
43 #if CONFIG_PGTABLE_LEVELS >= 5
44 #define p4d_ERROR(e)					\
45 	pr_err("%s:%d: bad p4d %p(%016lx)\n",		\
46 	       __FILE__, __LINE__, &(e), p4d_val(e))
47 #endif
48 
49 #define pgd_ERROR(e)					\
50 	pr_err("%s:%d: bad pgd %p(%016lx)\n",		\
51 	       __FILE__, __LINE__, &(e), pgd_val(e))
52 
53 struct mm_struct;
54 
55 void set_pte_vaddr_p4d(p4d_t *p4d_page, unsigned long vaddr, pte_t new_pte);
56 void set_pte_vaddr_pud(pud_t *pud_page, unsigned long vaddr, pte_t new_pte);
57 
58 static inline void native_pte_clear(struct mm_struct *mm, unsigned long addr,
59 				    pte_t *ptep)
60 {
61 	*ptep = native_make_pte(0);
62 }
63 
64 static inline void native_set_pte(pte_t *ptep, pte_t pte)
65 {
66 	*ptep = pte;
67 }
68 
69 static inline void native_set_pte_atomic(pte_t *ptep, pte_t pte)
70 {
71 	native_set_pte(ptep, pte);
72 }
73 
74 static inline void native_set_pmd(pmd_t *pmdp, pmd_t pmd)
75 {
76 	*pmdp = pmd;
77 }
78 
79 static inline void native_pmd_clear(pmd_t *pmd)
80 {
81 	native_set_pmd(pmd, native_make_pmd(0));
82 }
83 
84 static inline pte_t native_ptep_get_and_clear(pte_t *xp)
85 {
86 #ifdef CONFIG_SMP
87 	return native_make_pte(xchg(&xp->pte, 0));
88 #else
89 	/* native_local_ptep_get_and_clear,
90 	   but duplicated because of cyclic dependency */
91 	pte_t ret = *xp;
92 	native_pte_clear(NULL, 0, xp);
93 	return ret;
94 #endif
95 }
96 
97 static inline pmd_t native_pmdp_get_and_clear(pmd_t *xp)
98 {
99 #ifdef CONFIG_SMP
100 	return native_make_pmd(xchg(&xp->pmd, 0));
101 #else
102 	/* native_local_pmdp_get_and_clear,
103 	   but duplicated because of cyclic dependency */
104 	pmd_t ret = *xp;
105 	native_pmd_clear(xp);
106 	return ret;
107 #endif
108 }
109 
110 static inline void native_set_pud(pud_t *pudp, pud_t pud)
111 {
112 	*pudp = pud;
113 }
114 
115 static inline void native_pud_clear(pud_t *pud)
116 {
117 	native_set_pud(pud, native_make_pud(0));
118 }
119 
120 static inline pud_t native_pudp_get_and_clear(pud_t *xp)
121 {
122 #ifdef CONFIG_SMP
123 	return native_make_pud(xchg(&xp->pud, 0));
124 #else
125 	/* native_local_pudp_get_and_clear,
126 	 * but duplicated because of cyclic dependency
127 	 */
128 	pud_t ret = *xp;
129 
130 	native_pud_clear(xp);
131 	return ret;
132 #endif
133 }
134 
135 #ifdef CONFIG_PAGE_TABLE_ISOLATION
136 /*
137  * All top-level PAGE_TABLE_ISOLATION page tables are order-1 pages
138  * (8k-aligned and 8k in size).  The kernel one is at the beginning 4k and
139  * the user one is in the last 4k.  To switch between them, you
140  * just need to flip the 12th bit in their addresses.
141  */
142 #define PTI_PGTABLE_SWITCH_BIT	PAGE_SHIFT
143 
144 /*
145  * This generates better code than the inline assembly in
146  * __set_bit().
147  */
148 static inline void *ptr_set_bit(void *ptr, int bit)
149 {
150 	unsigned long __ptr = (unsigned long)ptr;
151 
152 	__ptr |= BIT(bit);
153 	return (void *)__ptr;
154 }
155 static inline void *ptr_clear_bit(void *ptr, int bit)
156 {
157 	unsigned long __ptr = (unsigned long)ptr;
158 
159 	__ptr &= ~BIT(bit);
160 	return (void *)__ptr;
161 }
162 
163 static inline pgd_t *kernel_to_user_pgdp(pgd_t *pgdp)
164 {
165 	return ptr_set_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
166 }
167 
168 static inline pgd_t *user_to_kernel_pgdp(pgd_t *pgdp)
169 {
170 	return ptr_clear_bit(pgdp, PTI_PGTABLE_SWITCH_BIT);
171 }
172 
173 static inline p4d_t *kernel_to_user_p4dp(p4d_t *p4dp)
174 {
175 	return ptr_set_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
176 }
177 
178 static inline p4d_t *user_to_kernel_p4dp(p4d_t *p4dp)
179 {
180 	return ptr_clear_bit(p4dp, PTI_PGTABLE_SWITCH_BIT);
181 }
182 #endif /* CONFIG_PAGE_TABLE_ISOLATION */
183 
184 /*
185  * Page table pages are page-aligned.  The lower half of the top
186  * level is used for userspace and the top half for the kernel.
187  *
188  * Returns true for parts of the PGD that map userspace and
189  * false for the parts that map the kernel.
190  */
191 static inline bool pgdp_maps_userspace(void *__ptr)
192 {
193 	unsigned long ptr = (unsigned long)__ptr;
194 
195 	return (ptr & ~PAGE_MASK) < (PAGE_SIZE / 2);
196 }
197 
198 #ifdef CONFIG_PAGE_TABLE_ISOLATION
199 pgd_t __pti_set_user_pgd(pgd_t *pgdp, pgd_t pgd);
200 
201 /*
202  * Take a PGD location (pgdp) and a pgd value that needs to be set there.
203  * Populates the user and returns the resulting PGD that must be set in
204  * the kernel copy of the page tables.
205  */
206 static inline pgd_t pti_set_user_pgd(pgd_t *pgdp, pgd_t pgd)
207 {
208 	if (!static_cpu_has(X86_FEATURE_PTI))
209 		return pgd;
210 	return __pti_set_user_pgd(pgdp, pgd);
211 }
212 #else
213 static inline pgd_t pti_set_user_pgd(pgd_t *pgdp, pgd_t pgd)
214 {
215 	return pgd;
216 }
217 #endif
218 
219 static inline void native_set_p4d(p4d_t *p4dp, p4d_t p4d)
220 {
221 	pgd_t pgd;
222 
223 	if (pgtable_l5_enabled || !IS_ENABLED(CONFIG_PAGE_TABLE_ISOLATION)) {
224 		*p4dp = p4d;
225 		return;
226 	}
227 
228 	pgd = native_make_pgd(native_p4d_val(p4d));
229 	pgd = pti_set_user_pgd((pgd_t *)p4dp, pgd);
230 	*p4dp = native_make_p4d(native_pgd_val(pgd));
231 }
232 
233 static inline void native_p4d_clear(p4d_t *p4d)
234 {
235 	native_set_p4d(p4d, native_make_p4d(0));
236 }
237 
238 static inline void native_set_pgd(pgd_t *pgdp, pgd_t pgd)
239 {
240 	*pgdp = pti_set_user_pgd(pgdp, pgd);
241 }
242 
243 static inline void native_pgd_clear(pgd_t *pgd)
244 {
245 	native_set_pgd(pgd, native_make_pgd(0));
246 }
247 
248 extern void sync_global_pgds(unsigned long start, unsigned long end);
249 
250 /*
251  * Conversion functions: convert a page and protection to a page entry,
252  * and a page entry and page directory to the page they refer to.
253  */
254 
255 /*
256  * Level 4 access.
257  */
258 static inline int pgd_large(pgd_t pgd) { return 0; }
259 #define mk_kernel_pgd(address) __pgd((address) | _KERNPG_TABLE)
260 
261 /* PUD - Level3 access */
262 
263 /* PMD  - Level 2 access */
264 
265 /* PTE - Level 1 access. */
266 
267 /* x86-64 always has all page tables mapped. */
268 #define pte_offset_map(dir, address) pte_offset_kernel((dir), (address))
269 #define pte_unmap(pte) ((void)(pte))/* NOP */
270 
271 /*
272  * Encode and de-code a swap entry
273  *
274  * |     ...            | 11| 10|  9|8|7|6|5| 4| 3|2| 1|0| <- bit number
275  * |     ...            |SW3|SW2|SW1|G|L|D|A|CD|WT|U| W|P| <- bit names
276  * | OFFSET (14->63) | TYPE (9-13)  |0|0|X|X| X| X|X|SD|0| <- swp entry
277  *
278  * G (8) is aliased and used as a PROT_NONE indicator for
279  * !present ptes.  We need to start storing swap entries above
280  * there.  We also need to avoid using A and D because of an
281  * erratum where they can be incorrectly set by hardware on
282  * non-present PTEs.
283  *
284  * SD (1) in swp entry is used to store soft dirty bit, which helps us
285  * remember soft dirty over page migration
286  *
287  * Bit 7 in swp entry should be 0 because pmd_present checks not only P,
288  * but also L and G.
289  */
290 #define SWP_TYPE_FIRST_BIT (_PAGE_BIT_PROTNONE + 1)
291 #define SWP_TYPE_BITS 5
292 /* Place the offset above the type: */
293 #define SWP_OFFSET_FIRST_BIT (SWP_TYPE_FIRST_BIT + SWP_TYPE_BITS)
294 
295 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS)
296 
297 #define __swp_type(x)			(((x).val >> (SWP_TYPE_FIRST_BIT)) \
298 					 & ((1U << SWP_TYPE_BITS) - 1))
299 #define __swp_offset(x)			((x).val >> SWP_OFFSET_FIRST_BIT)
300 #define __swp_entry(type, offset)	((swp_entry_t) { \
301 					 ((type) << (SWP_TYPE_FIRST_BIT)) \
302 					 | ((offset) << SWP_OFFSET_FIRST_BIT) })
303 #define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val((pte)) })
304 #define __pmd_to_swp_entry(pmd)		((swp_entry_t) { pmd_val((pmd)) })
305 #define __swp_entry_to_pte(x)		((pte_t) { .pte = (x).val })
306 #define __swp_entry_to_pmd(x)		((pmd_t) { .pmd = (x).val })
307 
308 extern int kern_addr_valid(unsigned long addr);
309 extern void cleanup_highmap(void);
310 
311 #define HAVE_ARCH_UNMAPPED_AREA
312 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
313 
314 #define pgtable_cache_init()   do { } while (0)
315 #define check_pgt_cache()      do { } while (0)
316 
317 #define PAGE_AGP    PAGE_KERNEL_NOCACHE
318 #define HAVE_PAGE_AGP 1
319 
320 /* fs/proc/kcore.c */
321 #define	kc_vaddr_to_offset(v) ((v) & __VIRTUAL_MASK)
322 #define	kc_offset_to_vaddr(o) ((o) | ~__VIRTUAL_MASK)
323 
324 #define __HAVE_ARCH_PTE_SAME
325 
326 #define vmemmap ((struct page *)VMEMMAP_START)
327 
328 extern void init_extra_mapping_uc(unsigned long phys, unsigned long size);
329 extern void init_extra_mapping_wb(unsigned long phys, unsigned long size);
330 
331 #define gup_fast_permitted gup_fast_permitted
332 static inline bool gup_fast_permitted(unsigned long start, int nr_pages,
333 		int write)
334 {
335 	unsigned long len, end;
336 
337 	len = (unsigned long)nr_pages << PAGE_SHIFT;
338 	end = start + len;
339 	if (end < start)
340 		return false;
341 	if (end >> __VIRTUAL_MASK_SHIFT)
342 		return false;
343 	return true;
344 }
345 
346 #endif /* !__ASSEMBLY__ */
347 #endif /* _ASM_X86_PGTABLE_64_H */
348