xref: /linux/arch/x86/include/asm/pgtable-3level.h (revision 8e07e0e3964ca4e23ce7b68e2096fe660a888942)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_X86_PGTABLE_3LEVEL_H
3 #define _ASM_X86_PGTABLE_3LEVEL_H
4 
5 /*
6  * Intel Physical Address Extension (PAE) Mode - three-level page
7  * tables on PPro+ CPUs.
8  *
9  * Copyright (C) 1999 Ingo Molnar <mingo@redhat.com>
10  */
11 
12 #define pte_ERROR(e)							\
13 	pr_err("%s:%d: bad pte %p(%08lx%08lx)\n",			\
14 	       __FILE__, __LINE__, &(e), (e).pte_high, (e).pte_low)
15 #define pmd_ERROR(e)							\
16 	pr_err("%s:%d: bad pmd %p(%016Lx)\n",				\
17 	       __FILE__, __LINE__, &(e), pmd_val(e))
18 #define pgd_ERROR(e)							\
19 	pr_err("%s:%d: bad pgd %p(%016Lx)\n",				\
20 	       __FILE__, __LINE__, &(e), pgd_val(e))
21 
22 #define pxx_xchg64(_pxx, _ptr, _val) ({					\
23 	_pxx##val_t *_p = (_pxx##val_t *)_ptr;				\
24 	_pxx##val_t _o = *_p;						\
25 	do { } while (!try_cmpxchg64(_p, &_o, (_val)));			\
26 	native_make_##_pxx(_o);						\
27 })
28 
29 /*
30  * Rules for using set_pte: the pte being assigned *must* be
31  * either not present or in a state where the hardware will
32  * not attempt to update the pte.  In places where this is
33  * not possible, use pte_get_and_clear to obtain the old pte
34  * value and then use set_pte to update it.  -ben
35  */
36 static inline void native_set_pte(pte_t *ptep, pte_t pte)
37 {
38 	WRITE_ONCE(ptep->pte_high, pte.pte_high);
39 	smp_wmb();
40 	WRITE_ONCE(ptep->pte_low, pte.pte_low);
41 }
42 
43 static inline void native_set_pte_atomic(pte_t *ptep, pte_t pte)
44 {
45 	pxx_xchg64(pte, ptep, native_pte_val(pte));
46 }
47 
48 static inline void native_set_pmd(pmd_t *pmdp, pmd_t pmd)
49 {
50 	pxx_xchg64(pmd, pmdp, native_pmd_val(pmd));
51 }
52 
53 static inline void native_set_pud(pud_t *pudp, pud_t pud)
54 {
55 #ifdef CONFIG_PAGE_TABLE_ISOLATION
56 	pud.p4d.pgd = pti_set_user_pgtbl(&pudp->p4d.pgd, pud.p4d.pgd);
57 #endif
58 	pxx_xchg64(pud, pudp, native_pud_val(pud));
59 }
60 
61 /*
62  * For PTEs and PDEs, we must clear the P-bit first when clearing a page table
63  * entry, so clear the bottom half first and enforce ordering with a compiler
64  * barrier.
65  */
66 static inline void native_pte_clear(struct mm_struct *mm, unsigned long addr,
67 				    pte_t *ptep)
68 {
69 	WRITE_ONCE(ptep->pte_low, 0);
70 	smp_wmb();
71 	WRITE_ONCE(ptep->pte_high, 0);
72 }
73 
74 static inline void native_pmd_clear(pmd_t *pmdp)
75 {
76 	WRITE_ONCE(pmdp->pmd_low, 0);
77 	smp_wmb();
78 	WRITE_ONCE(pmdp->pmd_high, 0);
79 }
80 
81 static inline void native_pud_clear(pud_t *pudp)
82 {
83 }
84 
85 static inline void pud_clear(pud_t *pudp)
86 {
87 	set_pud(pudp, __pud(0));
88 
89 	/*
90 	 * According to Intel App note "TLBs, Paging-Structure Caches,
91 	 * and Their Invalidation", April 2007, document 317080-001,
92 	 * section 8.1: in PAE mode we explicitly have to flush the
93 	 * TLB via cr3 if the top-level pgd is changed...
94 	 *
95 	 * Currently all places where pud_clear() is called either have
96 	 * flush_tlb_mm() followed or don't need TLB flush (x86_64 code or
97 	 * pud_clear_bad()), so we don't need TLB flush here.
98 	 */
99 }
100 
101 
102 #ifdef CONFIG_SMP
103 static inline pte_t native_ptep_get_and_clear(pte_t *ptep)
104 {
105 	return pxx_xchg64(pte, ptep, 0ULL);
106 }
107 
108 static inline pmd_t native_pmdp_get_and_clear(pmd_t *pmdp)
109 {
110 	return pxx_xchg64(pmd, pmdp, 0ULL);
111 }
112 
113 static inline pud_t native_pudp_get_and_clear(pud_t *pudp)
114 {
115 	return pxx_xchg64(pud, pudp, 0ULL);
116 }
117 #else
118 #define native_ptep_get_and_clear(xp) native_local_ptep_get_and_clear(xp)
119 #define native_pmdp_get_and_clear(xp) native_local_pmdp_get_and_clear(xp)
120 #define native_pudp_get_and_clear(xp) native_local_pudp_get_and_clear(xp)
121 #endif
122 
123 #ifndef pmdp_establish
124 #define pmdp_establish pmdp_establish
125 static inline pmd_t pmdp_establish(struct vm_area_struct *vma,
126 		unsigned long address, pmd_t *pmdp, pmd_t pmd)
127 {
128 	pmd_t old;
129 
130 	/*
131 	 * If pmd has present bit cleared we can get away without expensive
132 	 * cmpxchg64: we can update pmdp half-by-half without racing with
133 	 * anybody.
134 	 */
135 	if (!(pmd_val(pmd) & _PAGE_PRESENT)) {
136 		/* xchg acts as a barrier before setting of the high bits */
137 		old.pmd_low = xchg(&pmdp->pmd_low, pmd.pmd_low);
138 		old.pmd_high = READ_ONCE(pmdp->pmd_high);
139 		WRITE_ONCE(pmdp->pmd_high, pmd.pmd_high);
140 
141 		return old;
142 	}
143 
144 	return pxx_xchg64(pmd, pmdp, pmd.pmd);
145 }
146 #endif
147 
148 /*
149  * Encode/decode swap entries and swap PTEs. Swap PTEs are all PTEs that
150  * are !pte_none() && !pte_present().
151  *
152  * Format of swap PTEs:
153  *
154  *   6 6 6 6 5 5 5 5 5 5 5 5 5 5 4 4 4 4 4 4 4 4 4 4 3 3 3 3 3 3 3 3
155  *   3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2
156  *   < type -> <---------------------- offset ----------------------
157  *
158  *   3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
159  *   1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
160  *   --------------------------------------------> 0 E 0 0 0 0 0 0 0
161  *
162  *   E is the exclusive marker that is not stored in swap entries.
163  */
164 #define SWP_TYPE_BITS		5
165 #define _SWP_TYPE_MASK ((1U << SWP_TYPE_BITS) - 1)
166 
167 #define SWP_OFFSET_FIRST_BIT	(_PAGE_BIT_PROTNONE + 1)
168 
169 /* We always extract/encode the offset by shifting it all the way up, and then down again */
170 #define SWP_OFFSET_SHIFT	(SWP_OFFSET_FIRST_BIT + SWP_TYPE_BITS)
171 
172 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS)
173 #define __swp_type(x)			(((x).val) & _SWP_TYPE_MASK)
174 #define __swp_offset(x)			((x).val >> SWP_TYPE_BITS)
175 #define __swp_entry(type, offset)	((swp_entry_t){((type) & _SWP_TYPE_MASK) \
176 					| (offset) << SWP_TYPE_BITS})
177 
178 /*
179  * Normally, __swp_entry() converts from arch-independent swp_entry_t to
180  * arch-dependent swp_entry_t, and __swp_entry_to_pte() just stores the result
181  * to pte. But here we have 32bit swp_entry_t and 64bit pte, and need to use the
182  * whole 64 bits. Thus, we shift the "real" arch-dependent conversion to
183  * __swp_entry_to_pte() through the following helper macro based on 64bit
184  * __swp_entry().
185  */
186 #define __swp_pteval_entry(type, offset) ((pteval_t) { \
187 	(~(pteval_t)(offset) << SWP_OFFSET_SHIFT >> SWP_TYPE_BITS) \
188 	| ((pteval_t)(type) << (64 - SWP_TYPE_BITS)) })
189 
190 #define __swp_entry_to_pte(x)	((pte_t){ .pte = \
191 		__swp_pteval_entry(__swp_type(x), __swp_offset(x)) })
192 /*
193  * Analogically, __pte_to_swp_entry() doesn't just extract the arch-dependent
194  * swp_entry_t, but also has to convert it from 64bit to the 32bit
195  * intermediate representation, using the following macros based on 64bit
196  * __swp_type() and __swp_offset().
197  */
198 #define __pteval_swp_type(x) ((unsigned long)((x).pte >> (64 - SWP_TYPE_BITS)))
199 #define __pteval_swp_offset(x) ((unsigned long)(~((x).pte) << SWP_TYPE_BITS >> SWP_OFFSET_SHIFT))
200 
201 #define __pte_to_swp_entry(pte)	(__swp_entry(__pteval_swp_type(pte), \
202 					     __pteval_swp_offset(pte)))
203 
204 /* We borrow bit 7 to store the exclusive marker in swap PTEs. */
205 #define _PAGE_SWP_EXCLUSIVE	_PAGE_PSE
206 
207 #include <asm/pgtable-invert.h>
208 
209 #endif /* _ASM_X86_PGTABLE_3LEVEL_H */
210