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  * Define the address range of the kernel non-linear virtual area
31  */
32 
33 #ifdef CONFIG_PPC_BOOK3E
34 #define KERN_VIRT_START ASM_CONST(0x8000000000000000)
35 #else
36 #define KERN_VIRT_START ASM_CONST(0xD000000000000000)
37 #endif
38 #define KERN_VIRT_SIZE	ASM_CONST(0x0000100000000000)
39 
40 /*
41  * The vmalloc space starts at the beginning of that region, and
42  * occupies half of it on hash CPUs and a quarter of it on Book3E
43  * (we keep a quarter for the virtual memmap)
44  */
45 #define VMALLOC_START	KERN_VIRT_START
46 #ifdef CONFIG_PPC_BOOK3E
47 #define VMALLOC_SIZE	(KERN_VIRT_SIZE >> 2)
48 #else
49 #define VMALLOC_SIZE	(KERN_VIRT_SIZE >> 1)
50 #endif
51 #define VMALLOC_END	(VMALLOC_START + VMALLOC_SIZE)
52 
53 /*
54  * The second half of the kernel virtual space is used for IO mappings,
55  * it's itself carved into the PIO region (ISA and PHB IO space) and
56  * the ioremap space
57  *
58  *  ISA_IO_BASE = KERN_IO_START, 64K reserved area
59  *  PHB_IO_BASE = ISA_IO_BASE + 64K to ISA_IO_BASE + 2G, PHB IO spaces
60  * IOREMAP_BASE = ISA_IO_BASE + 2G to VMALLOC_START + PGTABLE_RANGE
61  */
62 #define KERN_IO_START	(KERN_VIRT_START + (KERN_VIRT_SIZE >> 1))
63 #define FULL_IO_SIZE	0x80000000ul
64 #define  ISA_IO_BASE	(KERN_IO_START)
65 #define  ISA_IO_END	(KERN_IO_START + 0x10000ul)
66 #define  PHB_IO_BASE	(ISA_IO_END)
67 #define  PHB_IO_END	(KERN_IO_START + FULL_IO_SIZE)
68 #define IOREMAP_BASE	(PHB_IO_END)
69 #define IOREMAP_END	(KERN_VIRT_START + KERN_VIRT_SIZE)
70 
71 
72 /*
73  * Region IDs
74  */
75 #define REGION_SHIFT		60UL
76 #define REGION_MASK		(0xfUL << REGION_SHIFT)
77 #define REGION_ID(ea)		(((unsigned long)(ea)) >> REGION_SHIFT)
78 
79 #define VMALLOC_REGION_ID	(REGION_ID(VMALLOC_START))
80 #define KERNEL_REGION_ID	(REGION_ID(PAGE_OFFSET))
81 #define VMEMMAP_REGION_ID	(0xfUL)	/* Server only */
82 #define USER_REGION_ID		(0UL)
83 
84 /*
85  * Defines the address of the vmemap area, in its own region on
86  * hash table CPUs and after the vmalloc space on Book3E
87  */
88 #ifdef CONFIG_PPC_BOOK3E
89 #define VMEMMAP_BASE		VMALLOC_END
90 #define VMEMMAP_END		KERN_IO_START
91 #else
92 #define VMEMMAP_BASE		(VMEMMAP_REGION_ID << REGION_SHIFT)
93 #endif
94 #define vmemmap			((struct page *)VMEMMAP_BASE)
95 
96 
97 /*
98  * Include the PTE bits definitions
99  */
100 #include <asm/nohash/pte-book3e.h>
101 #include <asm/pte-common.h>
102 
103 #ifdef CONFIG_PPC_MM_SLICES
104 #define HAVE_ARCH_UNMAPPED_AREA
105 #define HAVE_ARCH_UNMAPPED_AREA_TOPDOWN
106 #endif /* CONFIG_PPC_MM_SLICES */
107 
108 #ifndef __ASSEMBLY__
109 /* pte_clear moved to later in this file */
110 
111 /* Pointers in the page table tree are virtual addresses */
112 #define __pgtable_ptr_val(ptr)	((unsigned long)(ptr))
113 
114 #define PMD_BAD_BITS		(PTE_TABLE_SIZE-1)
115 #define PUD_BAD_BITS		(PMD_TABLE_SIZE-1)
116 
117 static inline void pmd_set(pmd_t *pmdp, unsigned long val)
118 {
119 	*pmdp = __pmd(val);
120 }
121 
122 static inline void pmd_clear(pmd_t *pmdp)
123 {
124 	*pmdp = __pmd(0);
125 }
126 
127 static inline pte_t pmd_pte(pmd_t pmd)
128 {
129 	return __pte(pmd_val(pmd));
130 }
131 
132 #define pmd_none(pmd)		(!pmd_val(pmd))
133 #define	pmd_bad(pmd)		(!is_kernel_addr(pmd_val(pmd)) \
134 				 || (pmd_val(pmd) & PMD_BAD_BITS))
135 #define	pmd_present(pmd)	(!pmd_none(pmd))
136 #define pmd_page_vaddr(pmd)	(pmd_val(pmd) & ~PMD_MASKED_BITS)
137 extern struct page *pmd_page(pmd_t pmd);
138 
139 static inline void pud_set(pud_t *pudp, unsigned long val)
140 {
141 	*pudp = __pud(val);
142 }
143 
144 static inline void pud_clear(pud_t *pudp)
145 {
146 	*pudp = __pud(0);
147 }
148 
149 #define pud_none(pud)		(!pud_val(pud))
150 #define	pud_bad(pud)		(!is_kernel_addr(pud_val(pud)) \
151 				 || (pud_val(pud) & PUD_BAD_BITS))
152 #define pud_present(pud)	(pud_val(pud) != 0)
153 #define pud_page_vaddr(pud)	(pud_val(pud) & ~PUD_MASKED_BITS)
154 
155 extern struct page *pud_page(pud_t pud);
156 
157 static inline pte_t pud_pte(pud_t pud)
158 {
159 	return __pte(pud_val(pud));
160 }
161 
162 static inline pud_t pte_pud(pte_t pte)
163 {
164 	return __pud(pte_val(pte));
165 }
166 #define pud_write(pud)		pte_write(pud_pte(pud))
167 #define pgd_write(pgd)		pte_write(pgd_pte(pgd))
168 
169 static inline void pgd_set(pgd_t *pgdp, unsigned long val)
170 {
171 	*pgdp = __pgd(val);
172 }
173 
174 /*
175  * Find an entry in a page-table-directory.  We combine the address region
176  * (the high order N bits) and the pgd portion of the address.
177  */
178 #define pgd_index(address) (((address) >> (PGDIR_SHIFT)) & (PTRS_PER_PGD - 1))
179 
180 #define pgd_offset(mm, address)	 ((mm)->pgd + pgd_index(address))
181 
182 #define pmd_offset(pudp,addr) \
183   (((pmd_t *) pud_page_vaddr(*(pudp))) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
184 
185 #define pte_offset_kernel(dir,addr) \
186   (((pte_t *) pmd_page_vaddr(*(dir))) + (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1)))
187 
188 #define pte_offset_map(dir,addr)	pte_offset_kernel((dir), (addr))
189 #define pte_unmap(pte)			do { } while(0)
190 
191 /* to find an entry in a kernel page-table-directory */
192 /* This now only contains the vmalloc pages */
193 #define pgd_offset_k(address) pgd_offset(&init_mm, address)
194 extern void hpte_need_flush(struct mm_struct *mm, unsigned long addr,
195 			    pte_t *ptep, unsigned long pte, int huge);
196 
197 /* Atomic PTE updates */
198 static inline unsigned long pte_update(struct mm_struct *mm,
199 				       unsigned long addr,
200 				       pte_t *ptep, unsigned long clr,
201 				       unsigned long set,
202 				       int huge)
203 {
204 #ifdef PTE_ATOMIC_UPDATES
205 	unsigned long old, tmp;
206 
207 	__asm__ __volatile__(
208 	"1:	ldarx	%0,0,%3		# pte_update\n\
209 	andi.	%1,%0,%6\n\
210 	bne-	1b \n\
211 	andc	%1,%0,%4 \n\
212 	or	%1,%1,%7\n\
213 	stdcx.	%1,0,%3 \n\
214 	bne-	1b"
215 	: "=&r" (old), "=&r" (tmp), "=m" (*ptep)
216 	: "r" (ptep), "r" (clr), "m" (*ptep), "i" (_PAGE_BUSY), "r" (set)
217 	: "cc" );
218 #else
219 	unsigned long old = pte_val(*ptep);
220 	*ptep = __pte((old & ~clr) | set);
221 #endif
222 	/* huge pages use the old page table lock */
223 	if (!huge)
224 		assert_pte_locked(mm, addr);
225 
226 #ifdef CONFIG_PPC_STD_MMU_64
227 	if (old & _PAGE_HASHPTE)
228 		hpte_need_flush(mm, addr, ptep, old, huge);
229 #endif
230 
231 	return old;
232 }
233 
234 static inline int __ptep_test_and_clear_young(struct mm_struct *mm,
235 					      unsigned long addr, pte_t *ptep)
236 {
237 	unsigned long old;
238 
239 	if ((pte_val(*ptep) & (_PAGE_ACCESSED | _PAGE_HASHPTE)) == 0)
240 		return 0;
241 	old = pte_update(mm, addr, ptep, _PAGE_ACCESSED, 0, 0);
242 	return (old & _PAGE_ACCESSED) != 0;
243 }
244 #define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
245 #define ptep_test_and_clear_young(__vma, __addr, __ptep)		   \
246 ({									   \
247 	int __r;							   \
248 	__r = __ptep_test_and_clear_young((__vma)->vm_mm, __addr, __ptep); \
249 	__r;								   \
250 })
251 
252 #define __HAVE_ARCH_PTEP_SET_WRPROTECT
253 static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr,
254 				      pte_t *ptep)
255 {
256 
257 	if ((pte_val(*ptep) & _PAGE_RW) == 0)
258 		return;
259 
260 	pte_update(mm, addr, ptep, _PAGE_RW, 0, 0);
261 }
262 
263 static inline void huge_ptep_set_wrprotect(struct mm_struct *mm,
264 					   unsigned long addr, pte_t *ptep)
265 {
266 	if ((pte_val(*ptep) & _PAGE_RW) == 0)
267 		return;
268 
269 	pte_update(mm, addr, ptep, _PAGE_RW, 0, 1);
270 }
271 
272 /*
273  * We currently remove entries from the hashtable regardless of whether
274  * the entry was young or dirty. The generic routines only flush if the
275  * entry was young or dirty which is not good enough.
276  *
277  * We should be more intelligent about this but for the moment we override
278  * these functions and force a tlb flush unconditionally
279  */
280 #define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
281 #define ptep_clear_flush_young(__vma, __address, __ptep)		\
282 ({									\
283 	int __young = __ptep_test_and_clear_young((__vma)->vm_mm, __address, \
284 						  __ptep);		\
285 	__young;							\
286 })
287 
288 #define __HAVE_ARCH_PTEP_GET_AND_CLEAR
289 static inline pte_t ptep_get_and_clear(struct mm_struct *mm,
290 				       unsigned long addr, pte_t *ptep)
291 {
292 	unsigned long old = pte_update(mm, addr, ptep, ~0UL, 0, 0);
293 	return __pte(old);
294 }
295 
296 static inline void pte_clear(struct mm_struct *mm, unsigned long addr,
297 			     pte_t * ptep)
298 {
299 	pte_update(mm, addr, ptep, ~0UL, 0, 0);
300 }
301 
302 
303 /* Set the dirty and/or accessed bits atomically in a linux PTE, this
304  * function doesn't need to flush the hash entry
305  */
306 static inline void __ptep_set_access_flags(pte_t *ptep, pte_t entry)
307 {
308 	unsigned long bits = pte_val(entry) &
309 		(_PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_RW | _PAGE_EXEC);
310 
311 #ifdef PTE_ATOMIC_UPDATES
312 	unsigned long old, tmp;
313 
314 	__asm__ __volatile__(
315 	"1:	ldarx	%0,0,%4\n\
316 		andi.	%1,%0,%6\n\
317 		bne-	1b \n\
318 		or	%0,%3,%0\n\
319 		stdcx.	%0,0,%4\n\
320 		bne-	1b"
321 	:"=&r" (old), "=&r" (tmp), "=m" (*ptep)
322 	:"r" (bits), "r" (ptep), "m" (*ptep), "i" (_PAGE_BUSY)
323 	:"cc");
324 #else
325 	unsigned long old = pte_val(*ptep);
326 	*ptep = __pte(old | bits);
327 #endif
328 }
329 
330 #define __HAVE_ARCH_PTE_SAME
331 #define pte_same(A,B)	(((pte_val(A) ^ pte_val(B)) & ~_PAGE_HPTEFLAGS) == 0)
332 
333 #define pte_ERROR(e) \
334 	pr_err("%s:%d: bad pte %08lx.\n", __FILE__, __LINE__, pte_val(e))
335 #define pmd_ERROR(e) \
336 	pr_err("%s:%d: bad pmd %08lx.\n", __FILE__, __LINE__, pmd_val(e))
337 #define pgd_ERROR(e) \
338 	pr_err("%s:%d: bad pgd %08lx.\n", __FILE__, __LINE__, pgd_val(e))
339 
340 /* Encode and de-code a swap entry */
341 #define MAX_SWAPFILES_CHECK() do { \
342 	BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > SWP_TYPE_BITS); \
343 	/*							\
344 	 * Don't have overlapping bits with _PAGE_HPTEFLAGS	\
345 	 * We filter HPTEFLAGS on set_pte.			\
346 	 */							\
347 	BUILD_BUG_ON(_PAGE_HPTEFLAGS & (0x1f << _PAGE_BIT_SWAP_TYPE)); \
348 	} while (0)
349 /*
350  * on pte we don't need handle RADIX_TREE_EXCEPTIONAL_SHIFT;
351  */
352 #define SWP_TYPE_BITS 5
353 #define __swp_type(x)		(((x).val >> _PAGE_BIT_SWAP_TYPE) \
354 				& ((1UL << SWP_TYPE_BITS) - 1))
355 #define __swp_offset(x)		((x).val >> PTE_RPN_SHIFT)
356 #define __swp_entry(type, offset)	((swp_entry_t) { \
357 					((type) << _PAGE_BIT_SWAP_TYPE) \
358 					| ((offset) << PTE_RPN_SHIFT) })
359 
360 #define __pte_to_swp_entry(pte)		((swp_entry_t) { pte_val((pte)) })
361 #define __swp_entry_to_pte(x)		__pte((x).val)
362 
363 void pgtable_cache_add(unsigned shift, void (*ctor)(void *));
364 void pgtable_cache_init(void);
365 #endif /* __ASSEMBLY__ */
366 
367 #endif /* _ASM_POWERPC_NOHASH_64_PGTABLE_H */
368