1#ifndef _I386_PGTABLE_H
2#define _I386_PGTABLE_H
3
4
5/*
6 * The Linux memory management assumes a three-level page table setup. On
7 * the i386, we use that, but "fold" the mid level into the top-level page
8 * table, so that we physically have the same two-level page table as the
9 * i386 mmu expects.
10 *
11 * This file contains the functions and defines necessary to modify and use
12 * the i386 page table tree.
13 */
14#ifndef __ASSEMBLY__
15#include <asm/processor.h>
16#include <asm/fixmap.h>
17#include <linux/threads.h>
18#include <asm/paravirt.h>
19
20#include <linux/bitops.h>
21#include <linux/slab.h>
22#include <linux/list.h>
23#include <linux/spinlock.h>
24
25struct mm_struct;
26struct vm_area_struct;
27
28/*
29 * ZERO_PAGE is a global shared page that is always zero: used
30 * for zero-mapped memory areas etc..
31 */
32#define ZERO_PAGE(vaddr) (virt_to_page(empty_zero_page))
33extern unsigned long empty_zero_page[1024];
34extern pgd_t swapper_pg_dir[1024];
35extern struct kmem_cache *pmd_cache;
36extern spinlock_t pgd_lock;
37extern struct page *pgd_list;
38void check_pgt_cache(void);
39
40void pmd_ctor(struct kmem_cache *, void *);
41void pgtable_cache_init(void);
42void paging_init(void);
43
44
45/*
46 * The Linux x86 paging architecture is 'compile-time dual-mode', it
47 * implements both the traditional 2-level x86 page tables and the
48 * newer 3-level PAE-mode page tables.
49 */
50#ifdef CONFIG_X86_PAE
51# include <asm/pgtable-3level-defs.h>
52# define PMD_SIZE	(1UL << PMD_SHIFT)
53# define PMD_MASK	(~(PMD_SIZE-1))
54#else
55# include <asm/pgtable-2level-defs.h>
56#endif
57
58#define PGDIR_SIZE	(1UL << PGDIR_SHIFT)
59#define PGDIR_MASK	(~(PGDIR_SIZE-1))
60
61#define USER_PTRS_PER_PGD	(TASK_SIZE/PGDIR_SIZE)
62#define FIRST_USER_ADDRESS	0
63
64#define USER_PGD_PTRS (PAGE_OFFSET >> PGDIR_SHIFT)
65#define KERNEL_PGD_PTRS (PTRS_PER_PGD-USER_PGD_PTRS)
66
67#define TWOLEVEL_PGDIR_SHIFT	22
68#define BOOT_USER_PGD_PTRS (__PAGE_OFFSET >> TWOLEVEL_PGDIR_SHIFT)
69#define BOOT_KERNEL_PGD_PTRS (1024-BOOT_USER_PGD_PTRS)
70
71/* Just any arbitrary offset to the start of the vmalloc VM area: the
72 * current 8MB value just means that there will be a 8MB "hole" after the
73 * physical memory until the kernel virtual memory starts.  That means that
74 * any out-of-bounds memory accesses will hopefully be caught.
75 * The vmalloc() routines leaves a hole of 4kB between each vmalloced
76 * area for the same reason. ;)
77 */
78#define VMALLOC_OFFSET	(8*1024*1024)
79#define VMALLOC_START	(((unsigned long) high_memory + \
80			2*VMALLOC_OFFSET-1) & ~(VMALLOC_OFFSET-1))
81#ifdef CONFIG_HIGHMEM
82# define VMALLOC_END	(PKMAP_BASE-2*PAGE_SIZE)
83#else
84# define VMALLOC_END	(FIXADDR_START-2*PAGE_SIZE)
85#endif
86
87/*
88 * _PAGE_PSE set in the page directory entry just means that
89 * the page directory entry points directly to a 4MB-aligned block of
90 * memory.
91 */
92#define _PAGE_BIT_PRESENT	0
93#define _PAGE_BIT_RW		1
94#define _PAGE_BIT_USER		2
95#define _PAGE_BIT_PWT		3
96#define _PAGE_BIT_PCD		4
97#define _PAGE_BIT_ACCESSED	5
98#define _PAGE_BIT_DIRTY		6
99#define _PAGE_BIT_PSE		7	/* 4 MB (or 2MB) page, Pentium+, if present.. */
100#define _PAGE_BIT_GLOBAL	8	/* Global TLB entry PPro+ */
101#define _PAGE_BIT_UNUSED1	9	/* available for programmer */
102#define _PAGE_BIT_UNUSED2	10
103#define _PAGE_BIT_UNUSED3	11
104#define _PAGE_BIT_NX		63
105
106#define _PAGE_PRESENT	0x001
107#define _PAGE_RW	0x002
108#define _PAGE_USER	0x004
109#define _PAGE_PWT	0x008
110#define _PAGE_PCD	0x010
111#define _PAGE_ACCESSED	0x020
112#define _PAGE_DIRTY	0x040
113#define _PAGE_PSE	0x080	/* 4 MB (or 2MB) page, Pentium+, if present.. */
114#define _PAGE_GLOBAL	0x100	/* Global TLB entry PPro+ */
115#define _PAGE_UNUSED1	0x200	/* available for programmer */
116#define _PAGE_UNUSED2	0x400
117#define _PAGE_UNUSED3	0x800
118
119/* If _PAGE_PRESENT is clear, we use these: */
120#define _PAGE_FILE	0x040	/* nonlinear file mapping, saved PTE; unset:swap */
121#define _PAGE_PROTNONE	0x080	/* if the user mapped it with PROT_NONE;
122				   pte_present gives true */
123#ifdef CONFIG_X86_PAE
124#define _PAGE_NX	(1ULL<<_PAGE_BIT_NX)
125#else
126#define _PAGE_NX	0
127#endif
128
129#define _PAGE_TABLE	(_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED | _PAGE_DIRTY)
130#define _KERNPG_TABLE	(_PAGE_PRESENT | _PAGE_RW | _PAGE_ACCESSED | _PAGE_DIRTY)
131#define _PAGE_CHG_MASK	(PTE_MASK | _PAGE_ACCESSED | _PAGE_DIRTY)
132
133#define PAGE_NONE \
134	__pgprot(_PAGE_PROTNONE | _PAGE_ACCESSED)
135#define PAGE_SHARED \
136	__pgprot(_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED)
137
138#define PAGE_SHARED_EXEC \
139	__pgprot(_PAGE_PRESENT | _PAGE_RW | _PAGE_USER | _PAGE_ACCESSED)
140#define PAGE_COPY_NOEXEC \
141	__pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED | _PAGE_NX)
142#define PAGE_COPY_EXEC \
143	__pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED)
144#define PAGE_COPY \
145	PAGE_COPY_NOEXEC
146#define PAGE_READONLY \
147	__pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED | _PAGE_NX)
148#define PAGE_READONLY_EXEC \
149	__pgprot(_PAGE_PRESENT | _PAGE_USER | _PAGE_ACCESSED)
150
151#define _PAGE_KERNEL \
152	(_PAGE_PRESENT | _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED | _PAGE_NX)
153#define _PAGE_KERNEL_EXEC \
154	(_PAGE_PRESENT | _PAGE_RW | _PAGE_DIRTY | _PAGE_ACCESSED)
155
156extern unsigned long long __PAGE_KERNEL, __PAGE_KERNEL_EXEC;
157#define __PAGE_KERNEL_RO		(__PAGE_KERNEL & ~_PAGE_RW)
158#define __PAGE_KERNEL_RX		(__PAGE_KERNEL_EXEC & ~_PAGE_RW)
159#define __PAGE_KERNEL_NOCACHE		(__PAGE_KERNEL | _PAGE_PCD)
160#define __PAGE_KERNEL_LARGE		(__PAGE_KERNEL | _PAGE_PSE)
161#define __PAGE_KERNEL_LARGE_EXEC	(__PAGE_KERNEL_EXEC | _PAGE_PSE)
162
163#define PAGE_KERNEL		__pgprot(__PAGE_KERNEL)
164#define PAGE_KERNEL_RO		__pgprot(__PAGE_KERNEL_RO)
165#define PAGE_KERNEL_EXEC	__pgprot(__PAGE_KERNEL_EXEC)
166#define PAGE_KERNEL_RX		__pgprot(__PAGE_KERNEL_RX)
167#define PAGE_KERNEL_NOCACHE	__pgprot(__PAGE_KERNEL_NOCACHE)
168#define PAGE_KERNEL_LARGE	__pgprot(__PAGE_KERNEL_LARGE)
169#define PAGE_KERNEL_LARGE_EXEC	__pgprot(__PAGE_KERNEL_LARGE_EXEC)
170
171/*
172 * The i386 can't do page protection for execute, and considers that
173 * the same are read. Also, write permissions imply read permissions.
174 * This is the closest we can get..
175 */
176#define __P000	PAGE_NONE
177#define __P001	PAGE_READONLY
178#define __P010	PAGE_COPY
179#define __P011	PAGE_COPY
180#define __P100	PAGE_READONLY_EXEC
181#define __P101	PAGE_READONLY_EXEC
182#define __P110	PAGE_COPY_EXEC
183#define __P111	PAGE_COPY_EXEC
184
185#define __S000	PAGE_NONE
186#define __S001	PAGE_READONLY
187#define __S010	PAGE_SHARED
188#define __S011	PAGE_SHARED
189#define __S100	PAGE_READONLY_EXEC
190#define __S101	PAGE_READONLY_EXEC
191#define __S110	PAGE_SHARED_EXEC
192#define __S111	PAGE_SHARED_EXEC
193
194/*
195 * Define this if things work differently on an i386 and an i486:
196 * it will (on an i486) warn about kernel memory accesses that are
197 * done without a 'access_ok(VERIFY_WRITE,..)'
198 */
199#undef TEST_ACCESS_OK
200
201/* The boot page tables (all created as a single array) */
202extern unsigned long pg0[];
203
204#define pte_present(x)	((x).pte_low & (_PAGE_PRESENT | _PAGE_PROTNONE))
205
206/* To avoid harmful races, pmd_none(x) should check only the lower when PAE */
207#define pmd_none(x)	(!(unsigned long)pmd_val(x))
208#define pmd_present(x)	(pmd_val(x) & _PAGE_PRESENT)
209#define	pmd_bad(x)	((pmd_val(x) & (~PAGE_MASK & ~_PAGE_USER)) != _KERNPG_TABLE)
210
211
212#define pages_to_mb(x) ((x) >> (20-PAGE_SHIFT))
213
214/*
215 * The following only work if pte_present() is true.
216 * Undefined behaviour if not..
217 */
218static inline int pte_dirty(pte_t pte)		{ return (pte).pte_low & _PAGE_DIRTY; }
219static inline int pte_young(pte_t pte)		{ return (pte).pte_low & _PAGE_ACCESSED; }
220static inline int pte_write(pte_t pte)		{ return (pte).pte_low & _PAGE_RW; }
221static inline int pte_huge(pte_t pte)		{ return (pte).pte_low & _PAGE_PSE; }
222
223/*
224 * The following only works if pte_present() is not true.
225 */
226static inline int pte_file(pte_t pte)		{ return (pte).pte_low & _PAGE_FILE; }
227
228static inline pte_t pte_mkclean(pte_t pte)	{ (pte).pte_low &= ~_PAGE_DIRTY; return pte; }
229static inline pte_t pte_mkold(pte_t pte)	{ (pte).pte_low &= ~_PAGE_ACCESSED; return pte; }
230static inline pte_t pte_wrprotect(pte_t pte)	{ (pte).pte_low &= ~_PAGE_RW; return pte; }
231static inline pte_t pte_mkdirty(pte_t pte)	{ (pte).pte_low |= _PAGE_DIRTY; return pte; }
232static inline pte_t pte_mkyoung(pte_t pte)	{ (pte).pte_low |= _PAGE_ACCESSED; return pte; }
233static inline pte_t pte_mkwrite(pte_t pte)	{ (pte).pte_low |= _PAGE_RW; return pte; }
234static inline pte_t pte_mkhuge(pte_t pte)	{ (pte).pte_low |= _PAGE_PSE; return pte; }
235
236#ifdef CONFIG_X86_PAE
237# include <asm/pgtable-3level.h>
238#else
239# include <asm/pgtable-2level.h>
240#endif
241
242#ifndef CONFIG_PARAVIRT
243/*
244 * Rules for using pte_update - it must be called after any PTE update which
245 * has not been done using the set_pte / clear_pte interfaces.  It is used by
246 * shadow mode hypervisors to resynchronize the shadow page tables.  Kernel PTE
247 * updates should either be sets, clears, or set_pte_atomic for P->P
248 * transitions, which means this hook should only be called for user PTEs.
249 * This hook implies a P->P protection or access change has taken place, which
250 * requires a subsequent TLB flush.  The notification can optionally be delayed
251 * until the TLB flush event by using the pte_update_defer form of the
252 * interface, but care must be taken to assure that the flush happens while
253 * still holding the same page table lock so that the shadow and primary pages
254 * do not become out of sync on SMP.
255 */
256#define pte_update(mm, addr, ptep)		do { } while (0)
257#define pte_update_defer(mm, addr, ptep)	do { } while (0)
258#endif
259
260/* local pte updates need not use xchg for locking */
261static inline pte_t native_local_ptep_get_and_clear(pte_t *ptep)
262{
263	pte_t res = *ptep;
264
265	/* Pure native function needs no input for mm, addr */
266	native_pte_clear(NULL, 0, ptep);
267	return res;
268}
269
270/*
271 * We only update the dirty/accessed state if we set
272 * the dirty bit by hand in the kernel, since the hardware
273 * will do the accessed bit for us, and we don't want to
274 * race with other CPU's that might be updating the dirty
275 * bit at the same time.
276 */
277#define  __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
278#define ptep_set_access_flags(vma, address, ptep, entry, dirty)		\
279({									\
280	int __changed = !pte_same(*(ptep), entry);			\
281	if (__changed && dirty) {					\
282		(ptep)->pte_low = (entry).pte_low;			\
283		pte_update_defer((vma)->vm_mm, (address), (ptep));	\
284		flush_tlb_page(vma, address);				\
285	}								\
286	__changed;							\
287})
288
289#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
290#define ptep_test_and_clear_young(vma, addr, ptep) ({			\
291	int __ret = 0;							\
292	if (pte_young(*(ptep)))						\
293		__ret = test_and_clear_bit(_PAGE_BIT_ACCESSED,		\
294						&(ptep)->pte_low);	\
295	if (__ret)							\
296		pte_update((vma)->vm_mm, addr, ptep);			\
297	__ret;								\
298})
299
300#define __HAVE_ARCH_PTEP_CLEAR_YOUNG_FLUSH
301#define ptep_clear_flush_young(vma, address, ptep)			\
302({									\
303	int __young;							\
304	__young = ptep_test_and_clear_young((vma), (address), (ptep));	\
305	if (__young)							\
306		flush_tlb_page(vma, address);				\
307	__young;							\
308})
309
310#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
311static inline pte_t ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
312{
313	pte_t pte = native_ptep_get_and_clear(ptep);
314	pte_update(mm, addr, ptep);
315	return pte;
316}
317
318#define __HAVE_ARCH_PTEP_GET_AND_CLEAR_FULL
319static inline pte_t ptep_get_and_clear_full(struct mm_struct *mm, unsigned long addr, pte_t *ptep, int full)
320{
321	pte_t pte;
322	if (full) {
323		/*
324		 * Full address destruction in progress; paravirt does not
325		 * care about updates and native needs no locking
326		 */
327		pte = native_local_ptep_get_and_clear(ptep);
328	} else {
329		pte = ptep_get_and_clear(mm, addr, ptep);
330	}
331	return pte;
332}
333
334#define __HAVE_ARCH_PTEP_SET_WRPROTECT
335static inline void ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
336{
337	clear_bit(_PAGE_BIT_RW, &ptep->pte_low);
338	pte_update(mm, addr, ptep);
339}
340
341/*
342 * clone_pgd_range(pgd_t *dst, pgd_t *src, int count);
343 *
344 *  dst - pointer to pgd range anwhere on a pgd page
345 *  src - ""
346 *  count - the number of pgds to copy.
347 *
348 * dst and src can be on the same page, but the range must not overlap,
349 * and must not cross a page boundary.
350 */
351static inline void clone_pgd_range(pgd_t *dst, pgd_t *src, int count)
352{
353       memcpy(dst, src, count * sizeof(pgd_t));
354}
355
356/*
357 * Macro to mark a page protection value as "uncacheable".  On processors which do not support
358 * it, this is a no-op.
359 */
360#define pgprot_noncached(prot)	((boot_cpu_data.x86 > 3)					  \
361				 ? (__pgprot(pgprot_val(prot) | _PAGE_PCD | _PAGE_PWT)) : (prot))
362
363/*
364 * Conversion functions: convert a page and protection to a page entry,
365 * and a page entry and page directory to the page they refer to.
366 */
367
368#define mk_pte(page, pgprot)	pfn_pte(page_to_pfn(page), (pgprot))
369
370static inline pte_t pte_modify(pte_t pte, pgprot_t newprot)
371{
372	pte.pte_low &= _PAGE_CHG_MASK;
373	pte.pte_low |= pgprot_val(newprot);
374#ifdef CONFIG_X86_PAE
375	/*
376	 * Chop off the NX bit (if present), and add the NX portion of
377	 * the newprot (if present):
378	 */
379	pte.pte_high &= ~(1 << (_PAGE_BIT_NX - 32));
380	pte.pte_high |= (pgprot_val(newprot) >> 32) & \
381					(__supported_pte_mask >> 32);
382#endif
383	return pte;
384}
385
386#define pmd_large(pmd) \
387((pmd_val(pmd) & (_PAGE_PSE|_PAGE_PRESENT)) == (_PAGE_PSE|_PAGE_PRESENT))
388
389/*
390 * the pgd page can be thought of an array like this: pgd_t[PTRS_PER_PGD]
391 *
392 * this macro returns the index of the entry in the pgd page which would
393 * control the given virtual address
394 */
395#define pgd_index(address) (((address) >> PGDIR_SHIFT) & (PTRS_PER_PGD-1))
396#define pgd_index_k(addr) pgd_index(addr)
397
398/*
399 * pgd_offset() returns a (pgd_t *)
400 * pgd_index() is used get the offset into the pgd page's array of pgd_t's;
401 */
402#define pgd_offset(mm, address) ((mm)->pgd+pgd_index(address))
403
404/*
405 * a shortcut which implies the use of the kernel's pgd, instead
406 * of a process's
407 */
408#define pgd_offset_k(address) pgd_offset(&init_mm, address)
409
410/*
411 * the pmd page can be thought of an array like this: pmd_t[PTRS_PER_PMD]
412 *
413 * this macro returns the index of the entry in the pmd page which would
414 * control the given virtual address
415 */
416#define pmd_index(address) \
417		(((address) >> PMD_SHIFT) & (PTRS_PER_PMD-1))
418
419/*
420 * the pte page can be thought of an array like this: pte_t[PTRS_PER_PTE]
421 *
422 * this macro returns the index of the entry in the pte page which would
423 * control the given virtual address
424 */
425#define pte_index(address) \
426		(((address) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
427#define pte_offset_kernel(dir, address) \
428	((pte_t *) pmd_page_vaddr(*(dir)) +  pte_index(address))
429
430#define pmd_page(pmd) (pfn_to_page(pmd_val(pmd) >> PAGE_SHIFT))
431
432#define pmd_page_vaddr(pmd) \
433		((unsigned long) __va(pmd_val(pmd) & PAGE_MASK))
434
435/*
436 * Helper function that returns the kernel pagetable entry controlling
437 * the virtual address 'address'. NULL means no pagetable entry present.
438 * NOTE: the return type is pte_t but if the pmd is PSE then we return it
439 * as a pte too.
440 */
441extern pte_t *lookup_address(unsigned long address);
442
443/*
444 * Make a given kernel text page executable/non-executable.
445 * Returns the previous executability setting of that page (which
446 * is used to restore the previous state). Used by the SMP bootup code.
447 * NOTE: this is an __init function for security reasons.
448 */
449#ifdef CONFIG_X86_PAE
450 extern int set_kernel_exec(unsigned long vaddr, int enable);
451#else
452 static inline int set_kernel_exec(unsigned long vaddr, int enable) { return 0;}
453#endif
454
455#if defined(CONFIG_HIGHPTE)
456#define pte_offset_map(dir, address) \
457	((pte_t *)kmap_atomic_pte(pmd_page(*(dir)),KM_PTE0) + pte_index(address))
458#define pte_offset_map_nested(dir, address) \
459	((pte_t *)kmap_atomic_pte(pmd_page(*(dir)),KM_PTE1) + pte_index(address))
460#define pte_unmap(pte) kunmap_atomic(pte, KM_PTE0)
461#define pte_unmap_nested(pte) kunmap_atomic(pte, KM_PTE1)
462#else
463#define pte_offset_map(dir, address) \
464	((pte_t *)page_address(pmd_page(*(dir))) + pte_index(address))
465#define pte_offset_map_nested(dir, address) pte_offset_map(dir, address)
466#define pte_unmap(pte) do { } while (0)
467#define pte_unmap_nested(pte) do { } while (0)
468#endif
469
470/* Clear a kernel PTE and flush it from the TLB */
471#define kpte_clear_flush(ptep, vaddr)					\
472do {									\
473	pte_clear(&init_mm, vaddr, ptep);				\
474	__flush_tlb_one(vaddr);						\
475} while (0)
476
477/*
478 * The i386 doesn't have any external MMU info: the kernel page
479 * tables contain all the necessary information.
480 */
481#define update_mmu_cache(vma,address,pte) do { } while (0)
482
483void native_pagetable_setup_start(pgd_t *base);
484void native_pagetable_setup_done(pgd_t *base);
485
486#ifndef CONFIG_PARAVIRT
487static inline void paravirt_pagetable_setup_start(pgd_t *base)
488{
489	native_pagetable_setup_start(base);
490}
491
492static inline void paravirt_pagetable_setup_done(pgd_t *base)
493{
494	native_pagetable_setup_done(base);
495}
496#endif	/* !CONFIG_PARAVIRT */
497
498#endif /* !__ASSEMBLY__ */
499
500#ifdef CONFIG_FLATMEM
501#define kern_addr_valid(addr)	(1)
502#endif /* CONFIG_FLATMEM */
503
504#define io_remap_pfn_range(vma, vaddr, pfn, size, prot)		\
505		remap_pfn_range(vma, vaddr, pfn, size, prot)
506
507#include <asm-generic/pgtable.h>
508
509#endif /* _I386_PGTABLE_H */
510