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 #ifdef CONFIG_PPC_MM_SLICES
92 #define HAVE_ARCH_UNMAPPED_AREA
93 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
94 #endif /* CONFIG_PPC_MM_SLICES */
95 
96 #ifndef __ASSEMBLY__
97 /* pte_clear moved to later in this file */
98 
99 #define PMD_BAD_BITS		(PTE_TABLE_SIZE-1)
100 #define PUD_BAD_BITS		(PMD_TABLE_SIZE-1)
101 
102 static inline void pmd_set(pmd_t *pmdp, unsigned long val)
103 {
104 	*pmdp = __pmd(val);
105 }
106 
107 static inline void pmd_clear(pmd_t *pmdp)
108 {
109 	*pmdp = __pmd(0);
110 }
111 
112 static inline pte_t pmd_pte(pmd_t pmd)
113 {
114 	return __pte(pmd_val(pmd));
115 }
116 
117 #define pmd_none(pmd)		(!pmd_val(pmd))
118 #define	pmd_bad(pmd)		(!is_kernel_addr(pmd_val(pmd)) \
119 				 || (pmd_val(pmd) & PMD_BAD_BITS))
120 #define	pmd_present(pmd)	(!pmd_none(pmd))
121 #define pmd_page_vaddr(pmd)	(pmd_val(pmd) & ~PMD_MASKED_BITS)
122 extern struct page *pmd_page(pmd_t pmd);
123 
124 static inline void pud_set(pud_t *pudp, unsigned long val)
125 {
126 	*pudp = __pud(val);
127 }
128 
129 static inline void pud_clear(pud_t *pudp)
130 {
131 	*pudp = __pud(0);
132 }
133 
134 #define pud_none(pud)		(!pud_val(pud))
135 #define	pud_bad(pud)		(!is_kernel_addr(pud_val(pud)) \
136 				 || (pud_val(pud) & PUD_BAD_BITS))
137 #define pud_present(pud)	(pud_val(pud) != 0)
138 #define pud_page_vaddr(pud)	(pud_val(pud) & ~PUD_MASKED_BITS)
139 
140 extern struct page *pud_page(pud_t pud);
141 
142 static inline pte_t pud_pte(pud_t pud)
143 {
144 	return __pte(pud_val(pud));
145 }
146 
147 static inline pud_t pte_pud(pte_t pte)
148 {
149 	return __pud(pte_val(pte));
150 }
151 #define pud_write(pud)		pte_write(pud_pte(pud))
152 #define pgd_write(pgd)		pte_write(pgd_pte(pgd))
153 
154 static inline void pgd_set(pgd_t *pgdp, unsigned long val)
155 {
156 	*pgdp = __pgd(val);
157 }
158 
159 /*
160  * Find an entry in a page-table-directory.  We combine the address region
161  * (the high order N bits) and the pgd portion of the address.
162  */
163 #define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & (PTRS_PER_PGD - 1))
164 
165 #define pgd_offset(mm, address)	 ((mm)->pgd + pgd_index(address))
166 
167 #define pmd_offset(pudp,addr) \
168   (((pmd_t *) pud_page_vaddr(*(pudp))) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
169 
170 #define pte_offset_kernel(dir,addr) \
171   (((pte_t *) pmd_page_vaddr(*(dir))) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
172 
173 #define pte_offset_map(dir,addr)	pte_offset_kernel((dir), (addr))
174 #define pte_unmap(pte)			do { } while(0)
175 
176 /* to find an entry in a kernel page-table-directory */
177 /* This now only contains the vmalloc pages */
178 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
179 extern void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
180 			    pte_t *ptep, unsigned long pte, int huge);
181 
182 /* Atomic PTE updates */
183 static inline unsigned long pte_update(struct mm_struct *mm,
184 				       unsigned long addr,
185 				       pte_t *ptep, unsigned long clr,
186 				       unsigned long set,
187 				       int huge)
188 {
189 #ifdef PTE_ATOMIC_UPDATES
190 	unsigned long old, tmp;
191 
192 	__asm__ __volatile__(
193 	"1:	ldarx	%0,0,%3		# pte_update\n\
194 	andi.	%1,%0,%6\n\
195 	bne-	1b \n\
196 	andc	%1,%0,%4 \n\
197 	or	%1,%1,%7\n\
198 	stdcx.	%1,0,%3 \n\
199 	bne-	1b"
200 	: "=&r" (old), "=&r" (tmp), "=m" (*ptep)
201 	: "r" (ptep), "r" (clr), "m" (*ptep), "i" (_PAGE_BUSY), "r" (set)
202 	: "cc" );
203 #else
204 	unsigned long old = pte_val(*ptep);
205 	*ptep = __pte((old & ~clr) | set);
206 #endif
207 	/* huge pages use the old page table lock */
208 	if (!huge)
209 		assert_pte_locked(mm, addr);
210 
211 #ifdef CONFIG_PPC_STD_MMU_64
212 	if (old & _PAGE_HASHPTE)
213 		hpte_need_flush(mm, addr, ptep, old, huge);
214 #endif
215 
216 	return old;
217 }
218 
219 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
220 					      unsigned long addr, pte_t *ptep)
221 {
222 	unsigned long old;
223 
224 	if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
225 		return 0;
226 	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
227 	return (old & _PAGE_ACCESSED) != 0;
228 }
229 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
230 #define ptep_test_and_clear_young(__vma, __addr, __ptep)		   \
231 ({									   \
232 	int __r;							   \
233 	__r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
234 	__r;								   \
235 })
236 
237 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
238 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
239 				      pte_t *ptep)
240 {
241 
242 	if ((pte_val(*ptep) & _PAGE_RW) == 0)
243 		return;
244 
245 	pte_update(mm, addr, ptep, _PAGE_RW, 0, 0);
246 }
247 
248 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
249 					   unsigned long addr, pte_t *ptep)
250 {
251 	if ((pte_val(*ptep) & _PAGE_RW) == 0)
252 		return;
253 
254 	pte_update(mm, addr, ptep, _PAGE_RW, 0, 1);
255 }
256 
257 /*
258  * We currently remove entries from the hashtable regardless of whether
259  * the entry was young or dirty. The generic routines only flush if the
260  * entry was young or dirty which is not good enough.
261  *
262  * We should be more intelligent about this but for the moment we override
263  * these functions and force a tlb flush unconditionally
264  */
265 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
266 #define ptep_clear_flush_young(__vma, __address, __ptep)		\
267 ({									\
268 	int __young = __ptep_test_and_clear_young((__vma)->vm_mm, __address, \
269 						  __ptep);		\
270 	__young;							\
271 })
272 
273 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
274 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
275 				       unsigned long addr, pte_t *ptep)
276 {
277 	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
278 	return __pte(old);
279 }
280 
281 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
282 			     pte_t * ptep)
283 {
284 	pte_update(mm, addr, ptep, ~0UL, 0, 0);
285 }
286 
287 
288 /* Set the dirty and/or accessed bits atomically in a linux PTE, this
289  * function doesn't need to flush the hash entry
290  */
291 static inline void __ptep_set_access_flags(struct mm_struct *mm,
292 					   pte_t *ptep, pte_t entry,
293 					   unsigned long address)
294 {
295 	unsigned long bits = pte_val(entry) &
296 		(_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_RW | _PAGE_EXEC);
297 
298 #ifdef PTE_ATOMIC_UPDATES
299 	unsigned long old, tmp;
300 
301 	__asm__ __volatile__(
302 	"1:	ldarx	%0,0,%4\n\
303 		andi.	%1,%0,%6\n\
304 		bne-	1b \n\
305 		or	%0,%3,%0\n\
306 		stdcx.	%0,0,%4\n\
307 		bne-	1b"
308 	:"=&r" (old), "=&r" (tmp), "=m" (*ptep)
309 	:"r" (bits), "r" (ptep), "m" (*ptep), "i" (_PAGE_BUSY)
310 	:"cc");
311 #else
312 	unsigned long old = pte_val(*ptep);
313 	*ptep = __pte(old | bits);
314 #endif
315 }
316 
317 #define __HAVE_ARCH_PTE_SAME
318 #define pte_same(A,B)	(((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
319 
320 #define pte_ERROR(e) \
321 	pr_err("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
322 #define pmd_ERROR(e) \
323 	pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
324 #define pgd_ERROR(e) \
325 	pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
326 
327 /* Encode and de-code a swap entry */
328 #define MAX_SWAPFILES_CHECK() do { \
329 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
330 	/*							\
331 	 * Don't have overlapping bits with _PAGE_HPTEFLAGS	\
332 	 * We filter HPTEFLAGS on set_pte.			\
333 	 */							\
334 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & (0x1f << _PAGE_BIT_SWAP_TYPE)); \
335 	} while (0)
336 /*
337  * on pte we don't need handle RADIX_TREE_EXCEPTIONAL_SHIFT;
338  */
339 #define SWP_TYPE_BITS 5
340 #define __swp_type(x)		(((x).val >> _PAGE_BIT_SWAP_TYPE) \
341 				& ((1UL << SWP_TYPE_BITS) - 1))
342 #define __swp_offset(x)		((x).val >> PTE_RPN_SHIFT)
343 #define __swp_entry(type, offset)	((swp_entry_t) { \
344 					((type) << _PAGE_BIT_SWAP_TYPE) \
345 					| ((offset) << PTE_RPN_SHIFT) })
346 
347 #define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val((pte)) })
348 #define __swp_entry_to_pte(x)		__pte((x).val)
349 
350 extern int map_kernel_page(unsigned long ea, unsigned long pa,
351 			   unsigned long flags);
352 extern int __meminit vmemmap_create_mapping(unsigned long start,
353 					    unsigned long page_size,
354 					    unsigned long phys);
355 extern void vmemmap_remove_mapping(unsigned long start,
356 				   unsigned long page_size);
357 #endif /* __ASSEMBLY__ */
358 
359 #endif /* _ASM_POWERPC_NOHASH_64_PGTABLE_H */
360