xref: /openbmc/linux/arch/arm/include/asm/pgtable.h (revision a09d2831)
1 /*
2  *  arch/arm/include/asm/pgtable.h
3  *
4  *  Copyright (C) 1995-2002 Russell King
5  *
6  * This program is free software; you can redistribute it and/or modify
7  * it under the terms of the GNU General Public License version 2 as
8  * published by the Free Software Foundation.
9  */
10 #ifndef _ASMARM_PGTABLE_H
11 #define _ASMARM_PGTABLE_H
12 
13 #include <asm-generic/4level-fixup.h>
14 #include <asm/proc-fns.h>
15 
16 #ifndef CONFIG_MMU
17 
18 #include "pgtable-nommu.h"
19 
20 #else
21 
22 #include <asm/memory.h>
23 #include <mach/vmalloc.h>
24 #include <asm/pgtable-hwdef.h>
25 
26 /*
27  * Just any arbitrary offset to the start of the vmalloc VM area: the
28  * current 8MB value just means that there will be a 8MB "hole" after the
29  * physical memory until the kernel virtual memory starts.  That means that
30  * any out-of-bounds memory accesses will hopefully be caught.
31  * The vmalloc() routines leaves a hole of 4kB between each vmalloced
32  * area for the same reason. ;)
33  *
34  * Note that platforms may override VMALLOC_START, but they must provide
35  * VMALLOC_END.  VMALLOC_END defines the (exclusive) limit of this space,
36  * which may not overlap IO space.
37  */
38 #ifndef VMALLOC_START
39 #define VMALLOC_OFFSET		(8*1024*1024)
40 #define VMALLOC_START		(((unsigned long)high_memory + VMALLOC_OFFSET) & ~(VMALLOC_OFFSET-1))
41 #endif
42 
43 /*
44  * Hardware-wise, we have a two level page table structure, where the first
45  * level has 4096 entries, and the second level has 256 entries.  Each entry
46  * is one 32-bit word.  Most of the bits in the second level entry are used
47  * by hardware, and there aren't any "accessed" and "dirty" bits.
48  *
49  * Linux on the other hand has a three level page table structure, which can
50  * be wrapped to fit a two level page table structure easily - using the PGD
51  * and PTE only.  However, Linux also expects one "PTE" table per page, and
52  * at least a "dirty" bit.
53  *
54  * Therefore, we tweak the implementation slightly - we tell Linux that we
55  * have 2048 entries in the first level, each of which is 8 bytes (iow, two
56  * hardware pointers to the second level.)  The second level contains two
57  * hardware PTE tables arranged contiguously, followed by Linux versions
58  * which contain the state information Linux needs.  We, therefore, end up
59  * with 512 entries in the "PTE" level.
60  *
61  * This leads to the page tables having the following layout:
62  *
63  *    pgd             pte
64  * |        |
65  * +--------+ +0
66  * |        |-----> +------------+ +0
67  * +- - - - + +4    |  h/w pt 0  |
68  * |        |-----> +------------+ +1024
69  * +--------+ +8    |  h/w pt 1  |
70  * |        |       +------------+ +2048
71  * +- - - - +       | Linux pt 0 |
72  * |        |       +------------+ +3072
73  * +--------+       | Linux pt 1 |
74  * |        |       +------------+ +4096
75  *
76  * See L_PTE_xxx below for definitions of bits in the "Linux pt", and
77  * PTE_xxx for definitions of bits appearing in the "h/w pt".
78  *
79  * PMD_xxx definitions refer to bits in the first level page table.
80  *
81  * The "dirty" bit is emulated by only granting hardware write permission
82  * iff the page is marked "writable" and "dirty" in the Linux PTE.  This
83  * means that a write to a clean page will cause a permission fault, and
84  * the Linux MM layer will mark the page dirty via handle_pte_fault().
85  * For the hardware to notice the permission change, the TLB entry must
86  * be flushed, and ptep_set_access_flags() does that for us.
87  *
88  * The "accessed" or "young" bit is emulated by a similar method; we only
89  * allow accesses to the page if the "young" bit is set.  Accesses to the
90  * page will cause a fault, and handle_pte_fault() will set the young bit
91  * for us as long as the page is marked present in the corresponding Linux
92  * PTE entry.  Again, ptep_set_access_flags() will ensure that the TLB is
93  * up to date.
94  *
95  * However, when the "young" bit is cleared, we deny access to the page
96  * by clearing the hardware PTE.  Currently Linux does not flush the TLB
97  * for us in this case, which means the TLB will retain the transation
98  * until either the TLB entry is evicted under pressure, or a context
99  * switch which changes the user space mapping occurs.
100  */
101 #define PTRS_PER_PTE		512
102 #define PTRS_PER_PMD		1
103 #define PTRS_PER_PGD		2048
104 
105 /*
106  * PMD_SHIFT determines the size of the area a second-level page table can map
107  * PGDIR_SHIFT determines what a third-level page table entry can map
108  */
109 #define PMD_SHIFT		21
110 #define PGDIR_SHIFT		21
111 
112 #define LIBRARY_TEXT_START	0x0c000000
113 
114 #ifndef __ASSEMBLY__
115 extern void __pte_error(const char *file, int line, unsigned long val);
116 extern void __pmd_error(const char *file, int line, unsigned long val);
117 extern void __pgd_error(const char *file, int line, unsigned long val);
118 
119 #define pte_ERROR(pte)		__pte_error(__FILE__, __LINE__, pte_val(pte))
120 #define pmd_ERROR(pmd)		__pmd_error(__FILE__, __LINE__, pmd_val(pmd))
121 #define pgd_ERROR(pgd)		__pgd_error(__FILE__, __LINE__, pgd_val(pgd))
122 #endif /* !__ASSEMBLY__ */
123 
124 #define PMD_SIZE		(1UL << PMD_SHIFT)
125 #define PMD_MASK		(~(PMD_SIZE-1))
126 #define PGDIR_SIZE		(1UL << PGDIR_SHIFT)
127 #define PGDIR_MASK		(~(PGDIR_SIZE-1))
128 
129 /*
130  * This is the lowest virtual address we can permit any user space
131  * mapping to be mapped at.  This is particularly important for
132  * non-high vector CPUs.
133  */
134 #define FIRST_USER_ADDRESS	PAGE_SIZE
135 
136 #define FIRST_USER_PGD_NR	1
137 #define USER_PTRS_PER_PGD	((TASK_SIZE/PGDIR_SIZE) - FIRST_USER_PGD_NR)
138 
139 /*
140  * section address mask and size definitions.
141  */
142 #define SECTION_SHIFT		20
143 #define SECTION_SIZE		(1UL << SECTION_SHIFT)
144 #define SECTION_MASK		(~(SECTION_SIZE-1))
145 
146 /*
147  * ARMv6 supersection address mask and size definitions.
148  */
149 #define SUPERSECTION_SHIFT	24
150 #define SUPERSECTION_SIZE	(1UL << SUPERSECTION_SHIFT)
151 #define SUPERSECTION_MASK	(~(SUPERSECTION_SIZE-1))
152 
153 /*
154  * "Linux" PTE definitions.
155  *
156  * We keep two sets of PTEs - the hardware and the linux version.
157  * This allows greater flexibility in the way we map the Linux bits
158  * onto the hardware tables, and allows us to have YOUNG and DIRTY
159  * bits.
160  *
161  * The PTE table pointer refers to the hardware entries; the "Linux"
162  * entries are stored 1024 bytes below.
163  */
164 #define L_PTE_PRESENT		(1 << 0)
165 #define L_PTE_YOUNG		(1 << 1)
166 #define L_PTE_FILE		(1 << 2)	/* only when !PRESENT */
167 #define L_PTE_DIRTY		(1 << 6)
168 #define L_PTE_WRITE		(1 << 7)
169 #define L_PTE_USER		(1 << 8)
170 #define L_PTE_EXEC		(1 << 9)
171 #define L_PTE_SHARED		(1 << 10)	/* shared(v6), coherent(xsc3) */
172 
173 /*
174  * These are the memory types, defined to be compatible with
175  * pre-ARMv6 CPUs cacheable and bufferable bits:   XXCB
176  */
177 #define L_PTE_MT_UNCACHED	(0x00 << 2)	/* 0000 */
178 #define L_PTE_MT_BUFFERABLE	(0x01 << 2)	/* 0001 */
179 #define L_PTE_MT_WRITETHROUGH	(0x02 << 2)	/* 0010 */
180 #define L_PTE_MT_WRITEBACK	(0x03 << 2)	/* 0011 */
181 #define L_PTE_MT_MINICACHE	(0x06 << 2)	/* 0110 (sa1100, xscale) */
182 #define L_PTE_MT_WRITEALLOC	(0x07 << 2)	/* 0111 */
183 #define L_PTE_MT_DEV_SHARED	(0x04 << 2)	/* 0100 */
184 #define L_PTE_MT_DEV_NONSHARED	(0x0c << 2)	/* 1100 */
185 #define L_PTE_MT_DEV_WC		(0x09 << 2)	/* 1001 */
186 #define L_PTE_MT_DEV_CACHED	(0x0b << 2)	/* 1011 */
187 #define L_PTE_MT_MASK		(0x0f << 2)
188 
189 #ifndef __ASSEMBLY__
190 
191 /*
192  * The pgprot_* and protection_map entries will be fixed up in runtime
193  * to include the cachable and bufferable bits based on memory policy,
194  * as well as any architecture dependent bits like global/ASID and SMP
195  * shared mapping bits.
196  */
197 #define _L_PTE_DEFAULT	L_PTE_PRESENT | L_PTE_YOUNG
198 
199 extern pgprot_t		pgprot_user;
200 extern pgprot_t		pgprot_kernel;
201 
202 #define _MOD_PROT(p, b)	__pgprot(pgprot_val(p) | (b))
203 
204 #define PAGE_NONE		pgprot_user
205 #define PAGE_SHARED		_MOD_PROT(pgprot_user, L_PTE_USER | L_PTE_WRITE)
206 #define PAGE_SHARED_EXEC	_MOD_PROT(pgprot_user, L_PTE_USER | L_PTE_WRITE | L_PTE_EXEC)
207 #define PAGE_COPY		_MOD_PROT(pgprot_user, L_PTE_USER)
208 #define PAGE_COPY_EXEC		_MOD_PROT(pgprot_user, L_PTE_USER | L_PTE_EXEC)
209 #define PAGE_READONLY		_MOD_PROT(pgprot_user, L_PTE_USER)
210 #define PAGE_READONLY_EXEC	_MOD_PROT(pgprot_user, L_PTE_USER | L_PTE_EXEC)
211 #define PAGE_KERNEL		pgprot_kernel
212 #define PAGE_KERNEL_EXEC	_MOD_PROT(pgprot_kernel, L_PTE_EXEC)
213 
214 #define __PAGE_NONE		__pgprot(_L_PTE_DEFAULT)
215 #define __PAGE_SHARED		__pgprot(_L_PTE_DEFAULT | L_PTE_USER | L_PTE_WRITE)
216 #define __PAGE_SHARED_EXEC	__pgprot(_L_PTE_DEFAULT | L_PTE_USER | L_PTE_WRITE | L_PTE_EXEC)
217 #define __PAGE_COPY		__pgprot(_L_PTE_DEFAULT | L_PTE_USER)
218 #define __PAGE_COPY_EXEC	__pgprot(_L_PTE_DEFAULT | L_PTE_USER | L_PTE_EXEC)
219 #define __PAGE_READONLY		__pgprot(_L_PTE_DEFAULT | L_PTE_USER)
220 #define __PAGE_READONLY_EXEC	__pgprot(_L_PTE_DEFAULT | L_PTE_USER | L_PTE_EXEC)
221 
222 #endif /* __ASSEMBLY__ */
223 
224 /*
225  * The table below defines the page protection levels that we insert into our
226  * Linux page table version.  These get translated into the best that the
227  * architecture can perform.  Note that on most ARM hardware:
228  *  1) We cannot do execute protection
229  *  2) If we could do execute protection, then read is implied
230  *  3) write implies read permissions
231  */
232 #define __P000  __PAGE_NONE
233 #define __P001  __PAGE_READONLY
234 #define __P010  __PAGE_COPY
235 #define __P011  __PAGE_COPY
236 #define __P100  __PAGE_READONLY_EXEC
237 #define __P101  __PAGE_READONLY_EXEC
238 #define __P110  __PAGE_COPY_EXEC
239 #define __P111  __PAGE_COPY_EXEC
240 
241 #define __S000  __PAGE_NONE
242 #define __S001  __PAGE_READONLY
243 #define __S010  __PAGE_SHARED
244 #define __S011  __PAGE_SHARED
245 #define __S100  __PAGE_READONLY_EXEC
246 #define __S101  __PAGE_READONLY_EXEC
247 #define __S110  __PAGE_SHARED_EXEC
248 #define __S111  __PAGE_SHARED_EXEC
249 
250 #ifndef __ASSEMBLY__
251 /*
252  * ZERO_PAGE is a global shared page that is always zero: used
253  * for zero-mapped memory areas etc..
254  */
255 extern struct page *empty_zero_page;
256 #define ZERO_PAGE(vaddr)	(empty_zero_page)
257 
258 #define pte_pfn(pte)		(pte_val(pte) >> PAGE_SHIFT)
259 #define pfn_pte(pfn,prot)	(__pte(((pfn) << PAGE_SHIFT) | pgprot_val(prot)))
260 
261 #define pte_none(pte)		(!pte_val(pte))
262 #define pte_clear(mm,addr,ptep)	set_pte_ext(ptep, __pte(0), 0)
263 #define pte_page(pte)		(pfn_to_page(pte_pfn(pte)))
264 #define pte_offset_kernel(dir,addr)	(pmd_page_vaddr(*(dir)) + __pte_index(addr))
265 
266 #define pte_offset_map(dir,addr)	(__pte_map(dir, KM_PTE0) + __pte_index(addr))
267 #define pte_offset_map_nested(dir,addr)	(__pte_map(dir, KM_PTE1) + __pte_index(addr))
268 #define pte_unmap(pte)			__pte_unmap(pte, KM_PTE0)
269 #define pte_unmap_nested(pte)		__pte_unmap(pte, KM_PTE1)
270 
271 #ifndef CONFIG_HIGHPTE
272 #define __pte_map(dir,km)	pmd_page_vaddr(*(dir))
273 #define __pte_unmap(pte,km)	do { } while (0)
274 #else
275 #define __pte_map(dir,km)	((pte_t *)kmap_atomic(pmd_page(*(dir)), km) + PTRS_PER_PTE)
276 #define __pte_unmap(pte,km)	kunmap_atomic((pte - PTRS_PER_PTE), km)
277 #endif
278 
279 #define set_pte_ext(ptep,pte,ext) cpu_set_pte_ext(ptep,pte,ext)
280 
281 #define set_pte_at(mm,addr,ptep,pteval) do { \
282 	set_pte_ext(ptep, pteval, (addr) >= TASK_SIZE ? 0 : PTE_EXT_NG); \
283  } while (0)
284 
285 /*
286  * The following only work if pte_present() is true.
287  * Undefined behaviour if not..
288  */
289 #define pte_present(pte)	(pte_val(pte) & L_PTE_PRESENT)
290 #define pte_write(pte)		(pte_val(pte) & L_PTE_WRITE)
291 #define pte_dirty(pte)		(pte_val(pte) & L_PTE_DIRTY)
292 #define pte_young(pte)		(pte_val(pte) & L_PTE_YOUNG)
293 #define pte_special(pte)	(0)
294 
295 #define PTE_BIT_FUNC(fn,op) \
296 static inline pte_t pte_##fn(pte_t pte) { pte_val(pte) op; return pte; }
297 
298 PTE_BIT_FUNC(wrprotect, &= ~L_PTE_WRITE);
299 PTE_BIT_FUNC(mkwrite,   |= L_PTE_WRITE);
300 PTE_BIT_FUNC(mkclean,   &= ~L_PTE_DIRTY);
301 PTE_BIT_FUNC(mkdirty,   |= L_PTE_DIRTY);
302 PTE_BIT_FUNC(mkold,     &= ~L_PTE_YOUNG);
303 PTE_BIT_FUNC(mkyoung,   |= L_PTE_YOUNG);
304 
305 static inline pte_t pte_mkspecial(pte_t pte) { return pte; }
306 
307 #define __pgprot_modify(prot,mask,bits)		\
308 	__pgprot((pgprot_val(prot) & ~(mask)) | (bits))
309 
310 /*
311  * Mark the prot value as uncacheable and unbufferable.
312  */
313 #define pgprot_noncached(prot) \
314 	__pgprot_modify(prot, L_PTE_MT_MASK, L_PTE_MT_UNCACHED)
315 #define pgprot_writecombine(prot) \
316 	__pgprot_modify(prot, L_PTE_MT_MASK, L_PTE_MT_BUFFERABLE)
317 #if __LINUX_ARM_ARCH__ >= 7
318 #define pgprot_dmacoherent(prot) \
319 	__pgprot_modify(prot, L_PTE_MT_MASK|L_PTE_EXEC, L_PTE_MT_BUFFERABLE)
320 #else
321 #define pgprot_dmacoherent(prot) \
322 	__pgprot_modify(prot, L_PTE_MT_MASK|L_PTE_EXEC, L_PTE_MT_UNCACHED)
323 #endif
324 
325 #define pmd_none(pmd)		(!pmd_val(pmd))
326 #define pmd_present(pmd)	(pmd_val(pmd))
327 #define pmd_bad(pmd)		(pmd_val(pmd) & 2)
328 
329 #define copy_pmd(pmdpd,pmdps)		\
330 	do {				\
331 		pmdpd[0] = pmdps[0];	\
332 		pmdpd[1] = pmdps[1];	\
333 		flush_pmd_entry(pmdpd);	\
334 	} while (0)
335 
336 #define pmd_clear(pmdp)			\
337 	do {				\
338 		pmdp[0] = __pmd(0);	\
339 		pmdp[1] = __pmd(0);	\
340 		clean_pmd_entry(pmdp);	\
341 	} while (0)
342 
343 static inline pte_t *pmd_page_vaddr(pmd_t pmd)
344 {
345 	unsigned long ptr;
346 
347 	ptr = pmd_val(pmd) & ~(PTRS_PER_PTE * sizeof(void *) - 1);
348 	ptr += PTRS_PER_PTE * sizeof(void *);
349 
350 	return __va(ptr);
351 }
352 
353 #define pmd_page(pmd)		pfn_to_page(__phys_to_pfn(pmd_val(pmd)))
354 
355 /*
356  * Conversion functions: convert a page and protection to a page entry,
357  * and a page entry and page directory to the page they refer to.
358  */
359 #define mk_pte(page,prot)	pfn_pte(page_to_pfn(page),prot)
360 
361 /*
362  * The "pgd_xxx()" functions here are trivial for a folded two-level
363  * setup: the pgd is never bad, and a pmd always exists (as it's folded
364  * into the pgd entry)
365  */
366 #define pgd_none(pgd)		(0)
367 #define pgd_bad(pgd)		(0)
368 #define pgd_present(pgd)	(1)
369 #define pgd_clear(pgdp)		do { } while (0)
370 #define set_pgd(pgd,pgdp)	do { } while (0)
371 
372 /* to find an entry in a page-table-directory */
373 #define pgd_index(addr)		((addr) >> PGDIR_SHIFT)
374 
375 #define pgd_offset(mm, addr)	((mm)->pgd+pgd_index(addr))
376 
377 /* to find an entry in a kernel page-table-directory */
378 #define pgd_offset_k(addr)	pgd_offset(&init_mm, addr)
379 
380 /* Find an entry in the second-level page table.. */
381 #define pmd_offset(dir, addr)	((pmd_t *)(dir))
382 
383 /* Find an entry in the third-level page table.. */
384 #define __pte_index(addr)	(((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
385 
386 static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
387 {
388 	const unsigned long mask = L_PTE_EXEC | L_PTE_WRITE | L_PTE_USER;
389 	pte_val(pte) = (pte_val(pte) & ~mask) | (pgprot_val(newprot) & mask);
390 	return pte;
391 }
392 
393 extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
394 
395 /*
396  * Encode and decode a swap entry.  Swap entries are stored in the Linux
397  * page tables as follows:
398  *
399  *   3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
400  *   1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
401  *   <--------------- offset --------------------> <- type --> 0 0 0
402  *
403  * This gives us up to 63 swap files and 32GB per swap file.  Note that
404  * the offset field is always non-zero.
405  */
406 #define __SWP_TYPE_SHIFT	3
407 #define __SWP_TYPE_BITS		6
408 #define __SWP_TYPE_MASK		((1 << __SWP_TYPE_BITS) - 1)
409 #define __SWP_OFFSET_SHIFT	(__SWP_TYPE_BITS + __SWP_TYPE_SHIFT)
410 
411 #define __swp_type(x)		(((x).val >> __SWP_TYPE_SHIFT) & __SWP_TYPE_MASK)
412 #define __swp_offset(x)		((x).val >> __SWP_OFFSET_SHIFT)
413 #define __swp_entry(type,offset) ((swp_entry_t) { ((type) << __SWP_TYPE_SHIFT) | ((offset) << __SWP_OFFSET_SHIFT) })
414 
415 #define __pte_to_swp_entry(pte)	((swp_entry_t) { pte_val(pte) })
416 #define __swp_entry_to_pte(swp)	((pte_t) { (swp).val })
417 
418 /*
419  * It is an error for the kernel to have more swap files than we can
420  * encode in the PTEs.  This ensures that we know when MAX_SWAPFILES
421  * is increased beyond what we presently support.
422  */
423 #define MAX_SWAPFILES_CHECK() BUILD_BUG_ON(MAX_SWAPFILES_SHIFT > __SWP_TYPE_BITS)
424 
425 /*
426  * Encode and decode a file entry.  File entries are stored in the Linux
427  * page tables as follows:
428  *
429  *   3 3 2 2 2 2 2 2 2 2 2 2 1 1 1 1 1 1 1 1 1 1
430  *   1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0 9 8 7 6 5 4 3 2 1 0
431  *   <----------------------- offset ------------------------> 1 0 0
432  */
433 #define pte_file(pte)		(pte_val(pte) & L_PTE_FILE)
434 #define pte_to_pgoff(x)		(pte_val(x) >> 3)
435 #define pgoff_to_pte(x)		__pte(((x) << 3) | L_PTE_FILE)
436 
437 #define PTE_FILE_MAX_BITS	29
438 
439 /* Needs to be defined here and not in linux/mm.h, as it is arch dependent */
440 /* FIXME: this is not correct */
441 #define kern_addr_valid(addr)	(1)
442 
443 #include <asm-generic/pgtable.h>
444 
445 /*
446  * We provide our own arch_get_unmapped_area to cope with VIPT caches.
447  */
448 #define HAVE_ARCH_UNMAPPED_AREA
449 
450 /*
451  * remap a physical page `pfn' of size `size' with page protection `prot'
452  * into virtual address `from'
453  */
454 #define io_remap_pfn_range(vma,from,pfn,size,prot) \
455 		remap_pfn_range(vma, from, pfn, size, prot)
456 
457 #define pgtable_cache_init() do { } while (0)
458 
459 #endif /* !__ASSEMBLY__ */
460 
461 #endif /* CONFIG_MMU */
462 
463 #endif /* _ASMARM_PGTABLE_H */
464