xref: /linux/arch/powerpc/include/asm/book3s/64/hash.h (revision e6a901a00822659181c93c86d8bbc2a17779fddc)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_POWERPC_BOOK3S_64_HASH_H
3 #define _ASM_POWERPC_BOOK3S_64_HASH_H
4 #ifdef __KERNEL__
5 
6 #include <asm/asm-const.h>
7 
8 /*
9  * Common bits between 4K and 64K pages in a linux-style PTE.
10  * Additional bits may be defined in pgtable-hash64-*.h
11  *
12  */
13 #define H_PTE_NONE_MASK		_PAGE_HPTEFLAGS
14 
15 #ifdef CONFIG_PPC_64K_PAGES
16 #include <asm/book3s/64/hash-64k.h>
17 #else
18 #include <asm/book3s/64/hash-4k.h>
19 #endif
20 
21 #define H_PTRS_PER_PTE		(1 << H_PTE_INDEX_SIZE)
22 #define H_PTRS_PER_PMD		(1 << H_PMD_INDEX_SIZE)
23 #define H_PTRS_PER_PUD		(1 << H_PUD_INDEX_SIZE)
24 
25 /* Bits to set in a PMD/PUD/PGD entry valid bit*/
26 #define HASH_PMD_VAL_BITS		(0x8000000000000000UL)
27 #define HASH_PUD_VAL_BITS		(0x8000000000000000UL)
28 #define HASH_PGD_VAL_BITS		(0x8000000000000000UL)
29 
30 /*
31  * Size of EA range mapped by our pagetables.
32  */
33 #define H_PGTABLE_EADDR_SIZE	(H_PTE_INDEX_SIZE + H_PMD_INDEX_SIZE + \
34 				 H_PUD_INDEX_SIZE + H_PGD_INDEX_SIZE + PAGE_SHIFT)
35 #define H_PGTABLE_RANGE		(ASM_CONST(1) << H_PGTABLE_EADDR_SIZE)
36 /*
37  * Top 2 bits are ignored in page table walk.
38  */
39 #define EA_MASK			(~(0xcUL << 60))
40 
41 /*
42  * We store the slot details in the second half of page table.
43  * Increase the pud level table so that hugetlb ptes can be stored
44  * at pud level.
45  */
46 #if defined(CONFIG_HUGETLB_PAGE) &&  defined(CONFIG_PPC_64K_PAGES)
47 #define H_PUD_CACHE_INDEX	(H_PUD_INDEX_SIZE + 1)
48 #else
49 #define H_PUD_CACHE_INDEX	(H_PUD_INDEX_SIZE)
50 #endif
51 
52 /*
53  * +------------------------------+
54  * |                              |
55  * |                              |
56  * |                              |
57  * +------------------------------+  Kernel virtual map end (0xc00e000000000000)
58  * |                              |
59  * |                              |
60  * |      512TB/16TB of vmemmap   |
61  * |                              |
62  * |                              |
63  * +------------------------------+  Kernel vmemmap  start
64  * |                              |
65  * |      512TB/16TB of IO map    |
66  * |                              |
67  * +------------------------------+  Kernel IO map start
68  * |                              |
69  * |      512TB/16TB of vmap      |
70  * |                              |
71  * +------------------------------+  Kernel virt start (0xc008000000000000)
72  * |                              |
73  * |                              |
74  * |                              |
75  * +------------------------------+  Kernel linear (0xc.....)
76  */
77 
78 #define H_VMALLOC_START		H_KERN_VIRT_START
79 #define H_VMALLOC_SIZE		H_KERN_MAP_SIZE
80 #define H_VMALLOC_END		(H_VMALLOC_START + H_VMALLOC_SIZE)
81 
82 #define H_KERN_IO_START		H_VMALLOC_END
83 #define H_KERN_IO_SIZE		H_KERN_MAP_SIZE
84 #define H_KERN_IO_END		(H_KERN_IO_START + H_KERN_IO_SIZE)
85 
86 #define H_VMEMMAP_START		H_KERN_IO_END
87 #define H_VMEMMAP_SIZE		H_KERN_MAP_SIZE
88 #define H_VMEMMAP_END		(H_VMEMMAP_START + H_VMEMMAP_SIZE)
89 
90 #define NON_LINEAR_REGION_ID(ea)	((((unsigned long)ea - H_KERN_VIRT_START) >> REGION_SHIFT) + 2)
91 
92 /*
93  * Region IDs
94  */
95 #define USER_REGION_ID		0
96 #define LINEAR_MAP_REGION_ID	1
97 #define VMALLOC_REGION_ID	NON_LINEAR_REGION_ID(H_VMALLOC_START)
98 #define IO_REGION_ID		NON_LINEAR_REGION_ID(H_KERN_IO_START)
99 #define VMEMMAP_REGION_ID	NON_LINEAR_REGION_ID(H_VMEMMAP_START)
100 #define INVALID_REGION_ID	(VMEMMAP_REGION_ID + 1)
101 
102 /*
103  * Defines the address of the vmemap area, in its own region on
104  * hash table CPUs.
105  */
106 
107 /* PTEIDX nibble */
108 #define _PTEIDX_SECONDARY	0x8
109 #define _PTEIDX_GROUP_IX	0x7
110 
111 #define H_PMD_BAD_BITS		(PTE_TABLE_SIZE-1)
112 #define H_PUD_BAD_BITS		(PMD_TABLE_SIZE-1)
113 
114 #ifndef __ASSEMBLY__
115 static inline int get_region_id(unsigned long ea)
116 {
117 	int region_id;
118 	int id = (ea >> 60UL);
119 
120 	if (id == 0)
121 		return USER_REGION_ID;
122 
123 	if (id != (PAGE_OFFSET >> 60))
124 		return INVALID_REGION_ID;
125 
126 	if (ea < H_KERN_VIRT_START)
127 		return LINEAR_MAP_REGION_ID;
128 
129 	BUILD_BUG_ON(NON_LINEAR_REGION_ID(H_VMALLOC_START) != 2);
130 
131 	region_id = NON_LINEAR_REGION_ID(ea);
132 	return region_id;
133 }
134 
135 static inline int hash__pmd_same(pmd_t pmd_a, pmd_t pmd_b)
136 {
137 	return (((pmd_raw(pmd_a) ^ pmd_raw(pmd_b)) & ~cpu_to_be64(_PAGE_HPTEFLAGS)) == 0);
138 }
139 
140 #define	hash__pmd_bad(pmd)		(pmd_val(pmd) & H_PMD_BAD_BITS)
141 
142 /*
143  * pud comparison that will work with both pte and page table pointer.
144  */
145 static inline int hash__pud_same(pud_t pud_a, pud_t pud_b)
146 {
147 	return (((pud_raw(pud_a) ^ pud_raw(pud_b)) & ~cpu_to_be64(_PAGE_HPTEFLAGS)) == 0);
148 }
149 #define	hash__pud_bad(pud)		(pud_val(pud) & H_PUD_BAD_BITS)
150 
151 static inline int hash__p4d_bad(p4d_t p4d)
152 {
153 	return (p4d_val(p4d) == 0);
154 }
155 #ifdef CONFIG_STRICT_KERNEL_RWX
156 extern void hash__mark_rodata_ro(void);
157 extern void hash__mark_initmem_nx(void);
158 #endif
159 
160 extern void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
161 			    pte_t *ptep, unsigned long pte, int huge);
162 unsigned long htab_convert_pte_flags(unsigned long pteflags, unsigned long flags);
163 /* Atomic PTE updates */
164 static inline unsigned long hash__pte_update(struct mm_struct *mm,
165 					 unsigned long addr,
166 					 pte_t *ptep, unsigned long clr,
167 					 unsigned long set,
168 					 int huge)
169 {
170 	__be64 old_be, tmp_be;
171 	unsigned long old;
172 
173 	__asm__ __volatile__(
174 	"1:	ldarx	%0,0,%3		# pte_update\n\
175 	and.	%1,%0,%6\n\
176 	bne-	1b \n\
177 	andc	%1,%0,%4 \n\
178 	or	%1,%1,%7\n\
179 	stdcx.	%1,0,%3 \n\
180 	bne-	1b"
181 	: "=&r" (old_be), "=&r" (tmp_be), "=m" (*ptep)
182 	: "r" (ptep), "r" (cpu_to_be64(clr)), "m" (*ptep),
183 	  "r" (cpu_to_be64(H_PAGE_BUSY)), "r" (cpu_to_be64(set))
184 	: "cc" );
185 	/* huge pages use the old page table lock */
186 	if (!huge)
187 		assert_pte_locked(mm, addr);
188 
189 	old = be64_to_cpu(old_be);
190 	if (old & H_PAGE_HASHPTE)
191 		hpte_need_flush(mm, addr, ptep, old, huge);
192 
193 	return old;
194 }
195 
196 /* Set the dirty and/or accessed bits atomically in a linux PTE, this
197  * function doesn't need to flush the hash entry
198  */
199 static inline void hash__ptep_set_access_flags(pte_t *ptep, pte_t entry)
200 {
201 	__be64 old, tmp, val, mask;
202 
203 	mask = cpu_to_be64(_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_READ | _PAGE_WRITE |
204 			   _PAGE_EXEC | _PAGE_SOFT_DIRTY);
205 
206 	val = pte_raw(entry) & mask;
207 
208 	__asm__ __volatile__(
209 	"1:	ldarx	%0,0,%4\n\
210 		and.	%1,%0,%6\n\
211 		bne-	1b \n\
212 		or	%0,%3,%0\n\
213 		stdcx.	%0,0,%4\n\
214 		bne-	1b"
215 	:"=&r" (old), "=&r" (tmp), "=m" (*ptep)
216 	:"r" (val), "r" (ptep), "m" (*ptep), "r" (cpu_to_be64(H_PAGE_BUSY))
217 	:"cc");
218 }
219 
220 static inline int hash__pte_same(pte_t pte_a, pte_t pte_b)
221 {
222 	return (((pte_raw(pte_a) ^ pte_raw(pte_b)) & ~cpu_to_be64(_PAGE_HPTEFLAGS)) == 0);
223 }
224 
225 static inline int hash__pte_none(pte_t pte)
226 {
227 	return (pte_val(pte) & ~H_PTE_NONE_MASK) == 0;
228 }
229 
230 unsigned long pte_get_hash_gslot(unsigned long vpn, unsigned long shift,
231 		int ssize, real_pte_t rpte, unsigned int subpg_index);
232 
233 /* This low level function performs the actual PTE insertion
234  * Setting the PTE depends on the MMU type and other factors. It's
235  * an horrible mess that I'm not going to try to clean up now but
236  * I'm keeping it in one place rather than spread around
237  */
238 static inline void hash__set_pte_at(struct mm_struct *mm, unsigned long addr,
239 				  pte_t *ptep, pte_t pte, int percpu)
240 {
241 	/*
242 	 * Anything else just stores the PTE normally. That covers all 64-bit
243 	 * cases, and 32-bit non-hash with 32-bit PTEs.
244 	 */
245 	*ptep = pte;
246 }
247 
248 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
249 extern void hpte_do_hugepage_flush(struct mm_struct *mm, unsigned long addr,
250 				   pmd_t *pmdp, unsigned long old_pmd);
251 #else
252 static inline void hpte_do_hugepage_flush(struct mm_struct *mm,
253 					  unsigned long addr, pmd_t *pmdp,
254 					  unsigned long old_pmd)
255 {
256 	WARN(1, "%s called with THP disabled\n", __func__);
257 }
258 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
259 
260 
261 int hash__map_kernel_page(unsigned long ea, unsigned long pa, pgprot_t prot);
262 extern int __meminit hash__vmemmap_create_mapping(unsigned long start,
263 					      unsigned long page_size,
264 					      unsigned long phys);
265 extern void hash__vmemmap_remove_mapping(unsigned long start,
266 				     unsigned long page_size);
267 
268 int hash__create_section_mapping(unsigned long start, unsigned long end,
269 				 int nid, pgprot_t prot);
270 int hash__remove_section_mapping(unsigned long start, unsigned long end);
271 
272 #endif /* !__ASSEMBLY__ */
273 #endif /* __KERNEL__ */
274 #endif /* _ASM_POWERPC_BOOK3S_64_HASH_H */
275