1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _ASM_POWERPC_NOHASH_64_PGTABLE_H
3 #define _ASM_POWERPC_NOHASH_64_PGTABLE_H
4 /*
5  * This file contains the functions and defines necessary to modify and use
6  * the ppc64 hashed page table.
7  */
8 
9 #ifdef CONFIG_PPC_64K_PAGES
10 #include <asm/nohash/64/pgtable-64k.h>
11 #else
12 #include <asm/nohash/64/pgtable-4k.h>
13 #endif
14 #include <asm/barrier.h>
15 
16 #define FIRST_USER_ADDRESS	0UL
17 
18 /*
19  * Size of EA range mapped by our pagetables.
20  */
21 #define PGTABLE_EADDR_SIZE (PTE_INDEX_SIZE + PMD_INDEX_SIZE + \
22 			    PUD_INDEX_SIZE + PGD_INDEX_SIZE + PAGE_SHIFT)
23 #define PGTABLE_RANGE (ASM_CONST(1) << PGTABLE_EADDR_SIZE)
24 
25 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
26 #define PMD_CACHE_INDEX	(PMD_INDEX_SIZE + 1)
27 #else
28 #define PMD_CACHE_INDEX	PMD_INDEX_SIZE
29 #endif
30 
31 /*
32  * Define the address range of the kernel non-linear virtual area
33  */
34 #define KERN_VIRT_START ASM_CONST(0x8000000000000000)
35 #define KERN_VIRT_SIZE	ASM_CONST(0x0000100000000000)
36 
37 /*
38  * The vmalloc space starts at the beginning of that region, and
39  * occupies half of it on hash CPUs and a quarter of it on Book3E
40  * (we keep a quarter for the virtual memmap)
41  */
42 #define VMALLOC_START	KERN_VIRT_START
43 #define VMALLOC_SIZE	(KERN_VIRT_SIZE >> 2)
44 #define VMALLOC_END	(VMALLOC_START + VMALLOC_SIZE)
45 
46 /*
47  * The second half of the kernel virtual space is used for IO mappings,
48  * it's itself carved into the PIO region (ISA and PHB IO space) and
49  * the ioremap space
50  *
51  *  ISA_IO_BASE = KERN_IO_START, 64K reserved area
52  *  PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
53  * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
54  */
55 #define KERN_IO_START	(KERN_VIRT_START + (KERN_VIRT_SIZE >> 1))
56 #define FULL_IO_SIZE	0x80000000ul
57 #define  ISA_IO_BASE	(KERN_IO_START)
58 #define  ISA_IO_END	(KERN_IO_START + 0x10000ul)
59 #define  PHB_IO_BASE	(ISA_IO_END)
60 #define  PHB_IO_END	(KERN_IO_START + FULL_IO_SIZE)
61 #define IOREMAP_BASE	(PHB_IO_END)
62 #define IOREMAP_END	(KERN_VIRT_START + KERN_VIRT_SIZE)
63 
64 
65 /*
66  * Region IDs
67  */
68 #define REGION_SHIFT		60UL
69 #define REGION_MASK		(0xfUL << REGION_SHIFT)
70 #define REGION_ID(ea)		(((unsigned long)(ea)) >> REGION_SHIFT)
71 
72 #define VMALLOC_REGION_ID	(REGION_ID(VMALLOC_START))
73 #define KERNEL_REGION_ID	(REGION_ID(PAGE_OFFSET))
74 #define VMEMMAP_REGION_ID	(0xfUL)	/* Server only */
75 #define USER_REGION_ID		(0UL)
76 
77 /*
78  * Defines the address of the vmemap area, in its own region on
79  * hash table CPUs and after the vmalloc space on Book3E
80  */
81 #define VMEMMAP_BASE		VMALLOC_END
82 #define VMEMMAP_END		KERN_IO_START
83 #define vmemmap			((struct page *)VMEMMAP_BASE)
84 
85 
86 /*
87  * Include the PTE bits definitions
88  */
89 #include <asm/nohash/pte-book3e.h>
90 #include <asm/pte-common.h>
91 
92 #ifndef __ASSEMBLY__
93 /* pte_clear moved to later in this file */
94 
95 #define PMD_BAD_BITS		(PTE_TABLE_SIZE-1)
96 #define PUD_BAD_BITS		(PMD_TABLE_SIZE-1)
97 
98 static inline void pmd_set(pmd_t *pmdp, unsigned long val)
99 {
100 	*pmdp = __pmd(val);
101 }
102 
103 static inline void pmd_clear(pmd_t *pmdp)
104 {
105 	*pmdp = __pmd(0);
106 }
107 
108 static inline pte_t pmd_pte(pmd_t pmd)
109 {
110 	return __pte(pmd_val(pmd));
111 }
112 
113 #define pmd_none(pmd)		(!pmd_val(pmd))
114 #define	pmd_bad(pmd)		(!is_kernel_addr(pmd_val(pmd)) \
115 				 || (pmd_val(pmd) & PMD_BAD_BITS))
116 #define	pmd_present(pmd)	(!pmd_none(pmd))
117 #define pmd_page_vaddr(pmd)	(pmd_val(pmd) & ~PMD_MASKED_BITS)
118 extern struct page *pmd_page(pmd_t pmd);
119 
120 static inline void pud_set(pud_t *pudp, unsigned long val)
121 {
122 	*pudp = __pud(val);
123 }
124 
125 static inline void pud_clear(pud_t *pudp)
126 {
127 	*pudp = __pud(0);
128 }
129 
130 #define pud_none(pud)		(!pud_val(pud))
131 #define	pud_bad(pud)		(!is_kernel_addr(pud_val(pud)) \
132 				 || (pud_val(pud) & PUD_BAD_BITS))
133 #define pud_present(pud)	(pud_val(pud) != 0)
134 #define pud_page_vaddr(pud)	(pud_val(pud) & ~PUD_MASKED_BITS)
135 
136 extern struct page *pud_page(pud_t pud);
137 
138 static inline pte_t pud_pte(pud_t pud)
139 {
140 	return __pte(pud_val(pud));
141 }
142 
143 static inline pud_t pte_pud(pte_t pte)
144 {
145 	return __pud(pte_val(pte));
146 }
147 #define pud_write(pud)		pte_write(pud_pte(pud))
148 #define pgd_write(pgd)		pte_write(pgd_pte(pgd))
149 
150 static inline void pgd_set(pgd_t *pgdp, unsigned long val)
151 {
152 	*pgdp = __pgd(val);
153 }
154 
155 /*
156  * Find an entry in a page-table-directory.  We combine the address region
157  * (the high order N bits) and the pgd portion of the address.
158  */
159 #define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & (PTRS_PER_PGD - 1))
160 
161 #define pgd_offset(mm, address)	 ((mm)->pgd + pgd_index(address))
162 
163 #define pmd_offset(pudp,addr) \
164   (((pmd_t *) pud_page_vaddr(*(pudp))) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
165 
166 #define pte_offset_kernel(dir,addr) \
167   (((pte_t *) pmd_page_vaddr(*(dir))) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
168 
169 #define pte_offset_map(dir,addr)	pte_offset_kernel((dir), (addr))
170 #define pte_unmap(pte)			do { } while(0)
171 
172 /* to find an entry in a kernel page-table-directory */
173 /* This now only contains the vmalloc pages */
174 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
175 extern void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
176 			    pte_t *ptep, unsigned long pte, int huge);
177 
178 /* Atomic PTE updates */
179 static inline unsigned long pte_update(struct mm_struct *mm,
180 				       unsigned long addr,
181 				       pte_t *ptep, unsigned long clr,
182 				       unsigned long set,
183 				       int huge)
184 {
185 #ifdef PTE_ATOMIC_UPDATES
186 	unsigned long old, tmp;
187 
188 	__asm__ __volatile__(
189 	"1:	ldarx	%0,0,%3		# pte_update\n\
190 	andi.	%1,%0,%6\n\
191 	bne-	1b \n\
192 	andc	%1,%0,%4 \n\
193 	or	%1,%1,%7\n\
194 	stdcx.	%1,0,%3 \n\
195 	bne-	1b"
196 	: "=&r" (old), "=&r" (tmp), "=m" (*ptep)
197 	: "r" (ptep), "r" (clr), "m" (*ptep), "i" (_PAGE_BUSY), "r" (set)
198 	: "cc" );
199 #else
200 	unsigned long old = pte_val(*ptep);
201 	*ptep = __pte((old & ~clr) | set);
202 #endif
203 	/* huge pages use the old page table lock */
204 	if (!huge)
205 		assert_pte_locked(mm, addr);
206 
207 #ifdef CONFIG_PPC_BOOK3S_64
208 	if (old & _PAGE_HASHPTE)
209 		hpte_need_flush(mm, addr, ptep, old, huge);
210 #endif
211 
212 	return old;
213 }
214 
215 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
216 					      unsigned long addr, pte_t *ptep)
217 {
218 	unsigned long old;
219 
220 	if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
221 		return 0;
222 	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
223 	return (old & _PAGE_ACCESSED) != 0;
224 }
225 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
226 #define ptep_test_and_clear_young(__vma, __addr, __ptep)		   \
227 ({									   \
228 	int __r;							   \
229 	__r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
230 	__r;								   \
231 })
232 
233 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
234 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
235 				      pte_t *ptep)
236 {
237 
238 	if ((pte_val(*ptep) & _PAGE_RW) == 0)
239 		return;
240 
241 	pte_update(mm, addr, ptep, _PAGE_RW, 0, 0);
242 }
243 
244 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
245 					   unsigned long addr, pte_t *ptep)
246 {
247 	if ((pte_val(*ptep) & _PAGE_RW) == 0)
248 		return;
249 
250 	pte_update(mm, addr, ptep, _PAGE_RW, 0, 1);
251 }
252 
253 /*
254  * We currently remove entries from the hashtable regardless of whether
255  * the entry was young or dirty. The generic routines only flush if the
256  * entry was young or dirty which is not good enough.
257  *
258  * We should be more intelligent about this but for the moment we override
259  * these functions and force a tlb flush unconditionally
260  */
261 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
262 #define ptep_clear_flush_young(__vma, __address, __ptep)		\
263 ({									\
264 	int __young = __ptep_test_and_clear_young((__vma)->vm_mm, __address, \
265 						  __ptep);		\
266 	__young;							\
267 })
268 
269 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
270 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
271 				       unsigned long addr, pte_t *ptep)
272 {
273 	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
274 	return __pte(old);
275 }
276 
277 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
278 			     pte_t * ptep)
279 {
280 	pte_update(mm, addr, ptep, ~0UL, 0, 0);
281 }
282 
283 
284 /* Set the dirty and/or accessed bits atomically in a linux PTE, this
285  * function doesn't need to flush the hash entry
286  */
287 static inline void __ptep_set_access_flags(struct mm_struct *mm,
288 					   pte_t *ptep, pte_t entry,
289 					   unsigned long address)
290 {
291 	unsigned long bits = pte_val(entry) &
292 		(_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_RW | _PAGE_EXEC);
293 
294 #ifdef PTE_ATOMIC_UPDATES
295 	unsigned long old, tmp;
296 
297 	__asm__ __volatile__(
298 	"1:	ldarx	%0,0,%4\n\
299 		andi.	%1,%0,%6\n\
300 		bne-	1b \n\
301 		or	%0,%3,%0\n\
302 		stdcx.	%0,0,%4\n\
303 		bne-	1b"
304 	:"=&r" (old), "=&r" (tmp), "=m" (*ptep)
305 	:"r" (bits), "r" (ptep), "m" (*ptep), "i" (_PAGE_BUSY)
306 	:"cc");
307 #else
308 	unsigned long old = pte_val(*ptep);
309 	*ptep = __pte(old | bits);
310 #endif
311 }
312 
313 #define __HAVE_ARCH_PTE_SAME
314 #define pte_same(A,B)	(((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
315 
316 #define pte_ERROR(e) \
317 	pr_err("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
318 #define pmd_ERROR(e) \
319 	pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
320 #define pgd_ERROR(e) \
321 	pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
322 
323 /* Encode and de-code a swap entry */
324 #define MAX_SWAPFILES_CHECK() do { \
325 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
326 	/*							\
327 	 * Don't have overlapping bits with _PAGE_HPTEFLAGS	\
328 	 * We filter HPTEFLAGS on set_pte.			\
329 	 */							\
330 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & (0x1f << _PAGE_BIT_SWAP_TYPE)); \
331 	} while (0)
332 /*
333  * on pte we don't need handle RADIX_TREE_EXCEPTIONAL_SHIFT;
334  */
335 #define SWP_TYPE_BITS 5
336 #define __swp_type(x)		(((x).val >> _PAGE_BIT_SWAP_TYPE) \
337 				& ((1UL << SWP_TYPE_BITS) - 1))
338 #define __swp_offset(x)		((x).val >> PTE_RPN_SHIFT)
339 #define __swp_entry(type, offset)	((swp_entry_t) { \
340 					((type) << _PAGE_BIT_SWAP_TYPE) \
341 					| ((offset) << PTE_RPN_SHIFT) })
342 
343 #define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val((pte)) })
344 #define __swp_entry_to_pte(x)		__pte((x).val)
345 
346 extern int map_kernel_page(unsigned long ea, unsigned long pa,
347 			   unsigned long flags);
348 extern int __meminit vmemmap_create_mapping(unsigned long start,
349 					    unsigned long page_size,
350 					    unsigned long phys);
351 extern void vmemmap_remove_mapping(unsigned long start,
352 				   unsigned long page_size);
353 #endif /* __ASSEMBLY__ */
354 
355 #endif /* _ASM_POWERPC_NOHASH_64_PGTABLE_H */
356