1 #ifndef _ASM_POWERPC_NOHASH_64_PGTABLE_H
2 #define _ASM_POWERPC_NOHASH_64_PGTABLE_H
3 /*
4  * This file contains the functions and defines necessary to modify and use
5  * the ppc64 hashed page table.
6  */
7 
8 #ifdef CONFIG_PPC_64K_PAGES
9 #include <asm/nohash/64/pgtable-64k.h>
10 #else
11 #include <asm/nohash/64/pgtable-4k.h>
12 #endif
13 #include <asm/barrier.h>
14 
15 #define FIRST_USER_ADDRESS	0UL
16 
17 /*
18  * Size of EA range mapped by our pagetables.
19  */
20 #define PGTABLE_EADDR_SIZE (PTE_INDEX_SIZE + PMD_INDEX_SIZE + \
21 			    PUD_INDEX_SIZE + PGD_INDEX_SIZE + PAGE_SHIFT)
22 #define PGTABLE_RANGE (ASM_CONST(1) << PGTABLE_EADDR_SIZE)
23 
24 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
25 #define PMD_CACHE_INDEX	(PMD_INDEX_SIZE + 1)
26 #else
27 #define PMD_CACHE_INDEX	PMD_INDEX_SIZE
28 #endif
29 
30 /*
31  * Define the address range of the kernel non-linear virtual area
32  */
33 #define KERN_VIRT_START ASM_CONST(0x8000000000000000)
34 #define KERN_VIRT_SIZE	ASM_CONST(0x0000100000000000)
35 
36 /*
37  * The vmalloc space starts at the beginning of that region, and
38  * occupies half of it on hash CPUs and a quarter of it on Book3E
39  * (we keep a quarter for the virtual memmap)
40  */
41 #define VMALLOC_START	KERN_VIRT_START
42 #define VMALLOC_SIZE	(KERN_VIRT_SIZE >> 2)
43 #define VMALLOC_END	(VMALLOC_START + VMALLOC_SIZE)
44 
45 /*
46  * The second half of the kernel virtual space is used for IO mappings,
47  * it's itself carved into the PIO region (ISA and PHB IO space) and
48  * the ioremap space
49  *
50  *  ISA_IO_BASE = KERN_IO_START, 64K reserved area
51  *  PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
52  * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
53  */
54 #define KERN_IO_START	(KERN_VIRT_START + (KERN_VIRT_SIZE >> 1))
55 #define FULL_IO_SIZE	0x80000000ul
56 #define  ISA_IO_BASE	(KERN_IO_START)
57 #define  ISA_IO_END	(KERN_IO_START + 0x10000ul)
58 #define  PHB_IO_BASE	(ISA_IO_END)
59 #define  PHB_IO_END	(KERN_IO_START + FULL_IO_SIZE)
60 #define IOREMAP_BASE	(PHB_IO_END)
61 #define IOREMAP_END	(KERN_VIRT_START + KERN_VIRT_SIZE)
62 
63 
64 /*
65  * Region IDs
66  */
67 #define REGION_SHIFT		60UL
68 #define REGION_MASK		(0xfUL << REGION_SHIFT)
69 #define REGION_ID(ea)		(((unsigned long)(ea)) >> REGION_SHIFT)
70 
71 #define VMALLOC_REGION_ID	(REGION_ID(VMALLOC_START))
72 #define KERNEL_REGION_ID	(REGION_ID(PAGE_OFFSET))
73 #define VMEMMAP_REGION_ID	(0xfUL)	/* Server only */
74 #define USER_REGION_ID		(0UL)
75 
76 /*
77  * Defines the address of the vmemap area, in its own region on
78  * hash table CPUs and after the vmalloc space on Book3E
79  */
80 #define VMEMMAP_BASE		VMALLOC_END
81 #define VMEMMAP_END		KERN_IO_START
82 #define vmemmap			((struct page *)VMEMMAP_BASE)
83 
84 
85 /*
86  * Include the PTE bits definitions
87  */
88 #include <asm/nohash/pte-book3e.h>
89 #include <asm/pte-common.h>
90 
91 #ifndef __ASSEMBLY__
92 /* pte_clear moved to later in this file */
93 
94 #define PMD_BAD_BITS		(PTE_TABLE_SIZE-1)
95 #define PUD_BAD_BITS		(PMD_TABLE_SIZE-1)
96 
97 static inline void pmd_set(pmd_t *pmdp, unsigned long val)
98 {
99 	*pmdp = __pmd(val);
100 }
101 
102 static inline void pmd_clear(pmd_t *pmdp)
103 {
104 	*pmdp = __pmd(0);
105 }
106 
107 static inline pte_t pmd_pte(pmd_t pmd)
108 {
109 	return __pte(pmd_val(pmd));
110 }
111 
112 #define pmd_none(pmd)		(!pmd_val(pmd))
113 #define	pmd_bad(pmd)		(!is_kernel_addr(pmd_val(pmd)) \
114 				 || (pmd_val(pmd) & PMD_BAD_BITS))
115 #define	pmd_present(pmd)	(!pmd_none(pmd))
116 #define pmd_page_vaddr(pmd)	(pmd_val(pmd) & ~PMD_MASKED_BITS)
117 extern struct page *pmd_page(pmd_t pmd);
118 
119 static inline void pud_set(pud_t *pudp, unsigned long val)
120 {
121 	*pudp = __pud(val);
122 }
123 
124 static inline void pud_clear(pud_t *pudp)
125 {
126 	*pudp = __pud(0);
127 }
128 
129 #define pud_none(pud)		(!pud_val(pud))
130 #define	pud_bad(pud)		(!is_kernel_addr(pud_val(pud)) \
131 				 || (pud_val(pud) & PUD_BAD_BITS))
132 #define pud_present(pud)	(pud_val(pud) != 0)
133 #define pud_page_vaddr(pud)	(pud_val(pud) & ~PUD_MASKED_BITS)
134 
135 extern struct page *pud_page(pud_t pud);
136 
137 static inline pte_t pud_pte(pud_t pud)
138 {
139 	return __pte(pud_val(pud));
140 }
141 
142 static inline pud_t pte_pud(pte_t pte)
143 {
144 	return __pud(pte_val(pte));
145 }
146 #define pud_write(pud)		pte_write(pud_pte(pud))
147 #define pgd_write(pgd)		pte_write(pgd_pte(pgd))
148 
149 static inline void pgd_set(pgd_t *pgdp, unsigned long val)
150 {
151 	*pgdp = __pgd(val);
152 }
153 
154 /*
155  * Find an entry in a page-table-directory.  We combine the address region
156  * (the high order N bits) and the pgd portion of the address.
157  */
158 #define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & (PTRS_PER_PGD - 1))
159 
160 #define pgd_offset(mm, address)	 ((mm)->pgd + pgd_index(address))
161 
162 #define pmd_offset(pudp,addr) \
163   (((pmd_t *) pud_page_vaddr(*(pudp))) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
164 
165 #define pte_offset_kernel(dir,addr) \
166   (((pte_t *) pmd_page_vaddr(*(dir))) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
167 
168 #define pte_offset_map(dir,addr)	pte_offset_kernel((dir), (addr))
169 #define pte_unmap(pte)			do { } while(0)
170 
171 /* to find an entry in a kernel page-table-directory */
172 /* This now only contains the vmalloc pages */
173 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
174 extern void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
175 			    pte_t *ptep, unsigned long pte, int huge);
176 
177 /* Atomic PTE updates */
178 static inline unsigned long pte_update(struct mm_struct *mm,
179 				       unsigned long addr,
180 				       pte_t *ptep, unsigned long clr,
181 				       unsigned long set,
182 				       int huge)
183 {
184 #ifdef PTE_ATOMIC_UPDATES
185 	unsigned long old, tmp;
186 
187 	__asm__ __volatile__(
188 	"1:	ldarx	%0,0,%3		# pte_update\n\
189 	andi.	%1,%0,%6\n\
190 	bne-	1b \n\
191 	andc	%1,%0,%4 \n\
192 	or	%1,%1,%7\n\
193 	stdcx.	%1,0,%3 \n\
194 	bne-	1b"
195 	: "=&r" (old), "=&r" (tmp), "=m" (*ptep)
196 	: "r" (ptep), "r" (clr), "m" (*ptep), "i" (_PAGE_BUSY), "r" (set)
197 	: "cc" );
198 #else
199 	unsigned long old = pte_val(*ptep);
200 	*ptep = __pte((old & ~clr) | set);
201 #endif
202 	/* huge pages use the old page table lock */
203 	if (!huge)
204 		assert_pte_locked(mm, addr);
205 
206 #ifdef CONFIG_PPC_STD_MMU_64
207 	if (old & _PAGE_HASHPTE)
208 		hpte_need_flush(mm, addr, ptep, old, huge);
209 #endif
210 
211 	return old;
212 }
213 
214 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
215 					      unsigned long addr, pte_t *ptep)
216 {
217 	unsigned long old;
218 
219 	if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
220 		return 0;
221 	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
222 	return (old & _PAGE_ACCESSED) != 0;
223 }
224 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
225 #define ptep_test_and_clear_young(__vma, __addr, __ptep)		   \
226 ({									   \
227 	int __r;							   \
228 	__r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
229 	__r;								   \
230 })
231 
232 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
233 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
234 				      pte_t *ptep)
235 {
236 
237 	if ((pte_val(*ptep) & _PAGE_RW) == 0)
238 		return;
239 
240 	pte_update(mm, addr, ptep, _PAGE_RW, 0, 0);
241 }
242 
243 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
244 					   unsigned long addr, pte_t *ptep)
245 {
246 	if ((pte_val(*ptep) & _PAGE_RW) == 0)
247 		return;
248 
249 	pte_update(mm, addr, ptep, _PAGE_RW, 0, 1);
250 }
251 
252 /*
253  * We currently remove entries from the hashtable regardless of whether
254  * the entry was young or dirty. The generic routines only flush if the
255  * entry was young or dirty which is not good enough.
256  *
257  * We should be more intelligent about this but for the moment we override
258  * these functions and force a tlb flush unconditionally
259  */
260 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
261 #define ptep_clear_flush_young(__vma, __address, __ptep)		\
262 ({									\
263 	int __young = __ptep_test_and_clear_young((__vma)->vm_mm, __address, \
264 						  __ptep);		\
265 	__young;							\
266 })
267 
268 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
269 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
270 				       unsigned long addr, pte_t *ptep)
271 {
272 	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
273 	return __pte(old);
274 }
275 
276 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
277 			     pte_t * ptep)
278 {
279 	pte_update(mm, addr, ptep, ~0UL, 0, 0);
280 }
281 
282 
283 /* Set the dirty and/or accessed bits atomically in a linux PTE, this
284  * function doesn't need to flush the hash entry
285  */
286 static inline void __ptep_set_access_flags(struct mm_struct *mm,
287 					   pte_t *ptep, pte_t entry,
288 					   unsigned long address)
289 {
290 	unsigned long bits = pte_val(entry) &
291 		(_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_RW | _PAGE_EXEC);
292 
293 #ifdef PTE_ATOMIC_UPDATES
294 	unsigned long old, tmp;
295 
296 	__asm__ __volatile__(
297 	"1:	ldarx	%0,0,%4\n\
298 		andi.	%1,%0,%6\n\
299 		bne-	1b \n\
300 		or	%0,%3,%0\n\
301 		stdcx.	%0,0,%4\n\
302 		bne-	1b"
303 	:"=&r" (old), "=&r" (tmp), "=m" (*ptep)
304 	:"r" (bits), "r" (ptep), "m" (*ptep), "i" (_PAGE_BUSY)
305 	:"cc");
306 #else
307 	unsigned long old = pte_val(*ptep);
308 	*ptep = __pte(old | bits);
309 #endif
310 }
311 
312 #define __HAVE_ARCH_PTE_SAME
313 #define pte_same(A,B)	(((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
314 
315 #define pte_ERROR(e) \
316 	pr_err("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
317 #define pmd_ERROR(e) \
318 	pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
319 #define pgd_ERROR(e) \
320 	pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
321 
322 /* Encode and de-code a swap entry */
323 #define MAX_SWAPFILES_CHECK() do { \
324 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
325 	/*							\
326 	 * Don't have overlapping bits with _PAGE_HPTEFLAGS	\
327 	 * We filter HPTEFLAGS on set_pte.			\
328 	 */							\
329 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & (0x1f << _PAGE_BIT_SWAP_TYPE)); \
330 	} while (0)
331 /*
332  * on pte we don't need handle RADIX_TREE_EXCEPTIONAL_SHIFT;
333  */
334 #define SWP_TYPE_BITS 5
335 #define __swp_type(x)		(((x).val >> _PAGE_BIT_SWAP_TYPE) \
336 				& ((1UL << SWP_TYPE_BITS) - 1))
337 #define __swp_offset(x)		((x).val >> PTE_RPN_SHIFT)
338 #define __swp_entry(type, offset)	((swp_entry_t) { \
339 					((type) << _PAGE_BIT_SWAP_TYPE) \
340 					| ((offset) << PTE_RPN_SHIFT) })
341 
342 #define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val((pte)) })
343 #define __swp_entry_to_pte(x)		__pte((x).val)
344 
345 extern int map_kernel_page(unsigned long ea, unsigned long pa,
346 			   unsigned long flags);
347 extern int __meminit vmemmap_create_mapping(unsigned long start,
348 					    unsigned long page_size,
349 					    unsigned long phys);
350 extern void vmemmap_remove_mapping(unsigned long start,
351 				   unsigned long page_size);
352 #endif /* __ASSEMBLY__ */
353 
354 #endif /* _ASM_POWERPC_NOHASH_64_PGTABLE_H */
355