Loading...
1#ifndef _ASM_IA64_PGTABLE_H
2#define _ASM_IA64_PGTABLE_H
3
4/*
5 * This file contains the functions and defines necessary to modify and use
6 * the IA-64 page table tree.
7 *
8 * This hopefully works with any (fixed) IA-64 page-size, as defined
9 * in <asm/page.h>.
10 *
11 * Copyright (C) 1998-2005 Hewlett-Packard Co
12 * David Mosberger-Tang <davidm@hpl.hp.com>
13 */
14
15
16#include <asm/mman.h>
17#include <asm/page.h>
18#include <asm/processor.h>
19#include <asm/types.h>
20
21#define IA64_MAX_PHYS_BITS 50 /* max. number of physical address bits (architected) */
22
23/*
24 * First, define the various bits in a PTE. Note that the PTE format
25 * matches the VHPT short format, the firt doubleword of the VHPD long
26 * format, and the first doubleword of the TLB insertion format.
27 */
28#define _PAGE_P_BIT 0
29#define _PAGE_A_BIT 5
30#define _PAGE_D_BIT 6
31
32#define _PAGE_P (1 << _PAGE_P_BIT) /* page present bit */
33#define _PAGE_MA_WB (0x0 << 2) /* write back memory attribute */
34#define _PAGE_MA_UC (0x4 << 2) /* uncacheable memory attribute */
35#define _PAGE_MA_UCE (0x5 << 2) /* UC exported attribute */
36#define _PAGE_MA_WC (0x6 << 2) /* write coalescing memory attribute */
37#define _PAGE_MA_NAT (0x7 << 2) /* not-a-thing attribute */
38#define _PAGE_MA_MASK (0x7 << 2)
39#define _PAGE_PL_0 (0 << 7) /* privilege level 0 (kernel) */
40#define _PAGE_PL_1 (1 << 7) /* privilege level 1 (unused) */
41#define _PAGE_PL_2 (2 << 7) /* privilege level 2 (unused) */
42#define _PAGE_PL_3 (3 << 7) /* privilege level 3 (user) */
43#define _PAGE_PL_MASK (3 << 7)
44#define _PAGE_AR_R (0 << 9) /* read only */
45#define _PAGE_AR_RX (1 << 9) /* read & execute */
46#define _PAGE_AR_RW (2 << 9) /* read & write */
47#define _PAGE_AR_RWX (3 << 9) /* read, write & execute */
48#define _PAGE_AR_R_RW (4 << 9) /* read / read & write */
49#define _PAGE_AR_RX_RWX (5 << 9) /* read & exec / read, write & exec */
50#define _PAGE_AR_RWX_RW (6 << 9) /* read, write & exec / read & write */
51#define _PAGE_AR_X_RX (7 << 9) /* exec & promote / read & exec */
52#define _PAGE_AR_MASK (7 << 9)
53#define _PAGE_AR_SHIFT 9
54#define _PAGE_A (1 << _PAGE_A_BIT) /* page accessed bit */
55#define _PAGE_D (1 << _PAGE_D_BIT) /* page dirty bit */
56#define _PAGE_PPN_MASK (((__IA64_UL(1) << IA64_MAX_PHYS_BITS) - 1) & ~0xfffUL)
57#define _PAGE_ED (__IA64_UL(1) << 52) /* exception deferral */
58#define _PAGE_PROTNONE (__IA64_UL(1) << 63)
59
60/* Valid only for a PTE with the present bit cleared: */
61#define _PAGE_FILE (1 << 1) /* see swap & file pte remarks below */
62
63#define _PFN_MASK _PAGE_PPN_MASK
64/* Mask of bits which may be changed by pte_modify(); the odd bits are there for _PAGE_PROTNONE */
65#define _PAGE_CHG_MASK (_PAGE_P | _PAGE_PROTNONE | _PAGE_PL_MASK | _PAGE_AR_MASK | _PAGE_ED)
66
67#define _PAGE_SIZE_4K 12
68#define _PAGE_SIZE_8K 13
69#define _PAGE_SIZE_16K 14
70#define _PAGE_SIZE_64K 16
71#define _PAGE_SIZE_256K 18
72#define _PAGE_SIZE_1M 20
73#define _PAGE_SIZE_4M 22
74#define _PAGE_SIZE_16M 24
75#define _PAGE_SIZE_64M 26
76#define _PAGE_SIZE_256M 28
77#define _PAGE_SIZE_1G 30
78#define _PAGE_SIZE_4G 32
79
80#define __ACCESS_BITS _PAGE_ED | _PAGE_A | _PAGE_P | _PAGE_MA_WB
81#define __DIRTY_BITS_NO_ED _PAGE_A | _PAGE_P | _PAGE_D | _PAGE_MA_WB
82#define __DIRTY_BITS _PAGE_ED | __DIRTY_BITS_NO_ED
83
84/*
85 * How many pointers will a page table level hold expressed in shift
86 */
87#define PTRS_PER_PTD_SHIFT (PAGE_SHIFT-3)
88
89/*
90 * Definitions for fourth level:
91 */
92#define PTRS_PER_PTE (__IA64_UL(1) << (PTRS_PER_PTD_SHIFT))
93
94/*
95 * Definitions for third level:
96 *
97 * PMD_SHIFT determines the size of the area a third-level page table
98 * can map.
99 */
100#define PMD_SHIFT (PAGE_SHIFT + (PTRS_PER_PTD_SHIFT))
101#define PMD_SIZE (1UL << PMD_SHIFT)
102#define PMD_MASK (~(PMD_SIZE-1))
103#define PTRS_PER_PMD (1UL << (PTRS_PER_PTD_SHIFT))
104
105#ifdef CONFIG_PGTABLE_4
106/*
107 * Definitions for second level:
108 *
109 * PUD_SHIFT determines the size of the area a second-level page table
110 * can map.
111 */
112#define PUD_SHIFT (PMD_SHIFT + (PTRS_PER_PTD_SHIFT))
113#define PUD_SIZE (1UL << PUD_SHIFT)
114#define PUD_MASK (~(PUD_SIZE-1))
115#define PTRS_PER_PUD (1UL << (PTRS_PER_PTD_SHIFT))
116#endif
117
118/*
119 * Definitions for first level:
120 *
121 * PGDIR_SHIFT determines what a first-level page table entry can map.
122 */
123#ifdef CONFIG_PGTABLE_4
124#define PGDIR_SHIFT (PUD_SHIFT + (PTRS_PER_PTD_SHIFT))
125#else
126#define PGDIR_SHIFT (PMD_SHIFT + (PTRS_PER_PTD_SHIFT))
127#endif
128#define PGDIR_SIZE (__IA64_UL(1) << PGDIR_SHIFT)
129#define PGDIR_MASK (~(PGDIR_SIZE-1))
130#define PTRS_PER_PGD_SHIFT PTRS_PER_PTD_SHIFT
131#define PTRS_PER_PGD (1UL << PTRS_PER_PGD_SHIFT)
132#define USER_PTRS_PER_PGD (5*PTRS_PER_PGD/8) /* regions 0-4 are user regions */
133#define FIRST_USER_ADDRESS 0
134
135/*
136 * All the normal masks have the "page accessed" bits on, as any time
137 * they are used, the page is accessed. They are cleared only by the
138 * page-out routines.
139 */
140#define PAGE_NONE __pgprot(_PAGE_PROTNONE | _PAGE_A)
141#define PAGE_SHARED __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RW)
142#define PAGE_READONLY __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_R)
143#define PAGE_COPY __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_R)
144#define PAGE_COPY_EXEC __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RX)
145#define PAGE_GATE __pgprot(__ACCESS_BITS | _PAGE_PL_0 | _PAGE_AR_X_RX)
146#define PAGE_KERNEL __pgprot(__DIRTY_BITS | _PAGE_PL_0 | _PAGE_AR_RWX)
147#define PAGE_KERNELRX __pgprot(__ACCESS_BITS | _PAGE_PL_0 | _PAGE_AR_RX)
148#define PAGE_KERNEL_UC __pgprot(__DIRTY_BITS | _PAGE_PL_0 | _PAGE_AR_RWX | \
149 _PAGE_MA_UC)
150
151# ifndef __ASSEMBLY__
152
153#include <linux/sched.h> /* for mm_struct */
154#include <linux/bitops.h>
155#include <asm/cacheflush.h>
156#include <asm/mmu_context.h>
157
158/*
159 * Next come the mappings that determine how mmap() protection bits
160 * (PROT_EXEC, PROT_READ, PROT_WRITE, PROT_NONE) get implemented. The
161 * _P version gets used for a private shared memory segment, the _S
162 * version gets used for a shared memory segment with MAP_SHARED on.
163 * In a private shared memory segment, we do a copy-on-write if a task
164 * attempts to write to the page.
165 */
166 /* xwr */
167#define __P000 PAGE_NONE
168#define __P001 PAGE_READONLY
169#define __P010 PAGE_READONLY /* write to priv pg -> copy & make writable */
170#define __P011 PAGE_READONLY /* ditto */
171#define __P100 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_X_RX)
172#define __P101 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RX)
173#define __P110 PAGE_COPY_EXEC
174#define __P111 PAGE_COPY_EXEC
175
176#define __S000 PAGE_NONE
177#define __S001 PAGE_READONLY
178#define __S010 PAGE_SHARED /* we don't have (and don't need) write-only */
179#define __S011 PAGE_SHARED
180#define __S100 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_X_RX)
181#define __S101 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RX)
182#define __S110 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RWX)
183#define __S111 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RWX)
184
185#define pgd_ERROR(e) printk("%s:%d: bad pgd %016lx.\n", __FILE__, __LINE__, pgd_val(e))
186#ifdef CONFIG_PGTABLE_4
187#define pud_ERROR(e) printk("%s:%d: bad pud %016lx.\n", __FILE__, __LINE__, pud_val(e))
188#endif
189#define pmd_ERROR(e) printk("%s:%d: bad pmd %016lx.\n", __FILE__, __LINE__, pmd_val(e))
190#define pte_ERROR(e) printk("%s:%d: bad pte %016lx.\n", __FILE__, __LINE__, pte_val(e))
191
192
193/*
194 * Some definitions to translate between mem_map, PTEs, and page addresses:
195 */
196
197
198/* Quick test to see if ADDR is a (potentially) valid physical address. */
199static inline long
200ia64_phys_addr_valid (unsigned long addr)
201{
202 return (addr & (local_cpu_data->unimpl_pa_mask)) == 0;
203}
204
205/*
206 * kern_addr_valid(ADDR) tests if ADDR is pointing to valid kernel
207 * memory. For the return value to be meaningful, ADDR must be >=
208 * PAGE_OFFSET. This operation can be relatively expensive (e.g.,
209 * require a hash-, or multi-level tree-lookup or something of that
210 * sort) but it guarantees to return TRUE only if accessing the page
211 * at that address does not cause an error. Note that there may be
212 * addresses for which kern_addr_valid() returns FALSE even though an
213 * access would not cause an error (e.g., this is typically true for
214 * memory mapped I/O regions.
215 *
216 * XXX Need to implement this for IA-64.
217 */
218#define kern_addr_valid(addr) (1)
219
220
221/*
222 * Now come the defines and routines to manage and access the three-level
223 * page table.
224 */
225
226
227#define VMALLOC_START (RGN_BASE(RGN_GATE) + 0x200000000UL)
228#ifdef CONFIG_VIRTUAL_MEM_MAP
229# define VMALLOC_END_INIT (RGN_BASE(RGN_GATE) + (1UL << (4*PAGE_SHIFT - 9)))
230extern unsigned long VMALLOC_END;
231#else
232#if defined(CONFIG_SPARSEMEM) && defined(CONFIG_SPARSEMEM_VMEMMAP)
233/* SPARSEMEM_VMEMMAP uses half of vmalloc... */
234# define VMALLOC_END (RGN_BASE(RGN_GATE) + (1UL << (4*PAGE_SHIFT - 10)))
235# define vmemmap ((struct page *)VMALLOC_END)
236#else
237# define VMALLOC_END (RGN_BASE(RGN_GATE) + (1UL << (4*PAGE_SHIFT - 9)))
238#endif
239#endif
240
241/* fs/proc/kcore.c */
242#define kc_vaddr_to_offset(v) ((v) - RGN_BASE(RGN_GATE))
243#define kc_offset_to_vaddr(o) ((o) + RGN_BASE(RGN_GATE))
244
245#define RGN_MAP_SHIFT (PGDIR_SHIFT + PTRS_PER_PGD_SHIFT - 3)
246#define RGN_MAP_LIMIT ((1UL << RGN_MAP_SHIFT) - PAGE_SIZE) /* per region addr limit */
247
248/*
249 * Conversion functions: convert page frame number (pfn) and a protection value to a page
250 * table entry (pte).
251 */
252#define pfn_pte(pfn, pgprot) \
253({ pte_t __pte; pte_val(__pte) = ((pfn) << PAGE_SHIFT) | pgprot_val(pgprot); __pte; })
254
255/* Extract pfn from pte. */
256#define pte_pfn(_pte) ((pte_val(_pte) & _PFN_MASK) >> PAGE_SHIFT)
257
258#define mk_pte(page, pgprot) pfn_pte(page_to_pfn(page), (pgprot))
259
260/* This takes a physical page address that is used by the remapping functions */
261#define mk_pte_phys(physpage, pgprot) \
262({ pte_t __pte; pte_val(__pte) = physpage + pgprot_val(pgprot); __pte; })
263
264#define pte_modify(_pte, newprot) \
265 (__pte((pte_val(_pte) & ~_PAGE_CHG_MASK) | (pgprot_val(newprot) & _PAGE_CHG_MASK)))
266
267#define pte_none(pte) (!pte_val(pte))
268#define pte_present(pte) (pte_val(pte) & (_PAGE_P | _PAGE_PROTNONE))
269#define pte_clear(mm,addr,pte) (pte_val(*(pte)) = 0UL)
270/* pte_page() returns the "struct page *" corresponding to the PTE: */
271#define pte_page(pte) virt_to_page(((pte_val(pte) & _PFN_MASK) + PAGE_OFFSET))
272
273#define pmd_none(pmd) (!pmd_val(pmd))
274#define pmd_bad(pmd) (!ia64_phys_addr_valid(pmd_val(pmd)))
275#define pmd_present(pmd) (pmd_val(pmd) != 0UL)
276#define pmd_clear(pmdp) (pmd_val(*(pmdp)) = 0UL)
277#define pmd_page_vaddr(pmd) ((unsigned long) __va(pmd_val(pmd) & _PFN_MASK))
278#define pmd_page(pmd) virt_to_page((pmd_val(pmd) + PAGE_OFFSET))
279
280#define pud_none(pud) (!pud_val(pud))
281#define pud_bad(pud) (!ia64_phys_addr_valid(pud_val(pud)))
282#define pud_present(pud) (pud_val(pud) != 0UL)
283#define pud_clear(pudp) (pud_val(*(pudp)) = 0UL)
284#define pud_page_vaddr(pud) ((unsigned long) __va(pud_val(pud) & _PFN_MASK))
285#define pud_page(pud) virt_to_page((pud_val(pud) + PAGE_OFFSET))
286
287#ifdef CONFIG_PGTABLE_4
288#define pgd_none(pgd) (!pgd_val(pgd))
289#define pgd_bad(pgd) (!ia64_phys_addr_valid(pgd_val(pgd)))
290#define pgd_present(pgd) (pgd_val(pgd) != 0UL)
291#define pgd_clear(pgdp) (pgd_val(*(pgdp)) = 0UL)
292#define pgd_page_vaddr(pgd) ((unsigned long) __va(pgd_val(pgd) & _PFN_MASK))
293#define pgd_page(pgd) virt_to_page((pgd_val(pgd) + PAGE_OFFSET))
294#endif
295
296/*
297 * The following have defined behavior only work if pte_present() is true.
298 */
299#define pte_write(pte) ((unsigned) (((pte_val(pte) & _PAGE_AR_MASK) >> _PAGE_AR_SHIFT) - 2) <= 4)
300#define pte_exec(pte) ((pte_val(pte) & _PAGE_AR_RX) != 0)
301#define pte_dirty(pte) ((pte_val(pte) & _PAGE_D) != 0)
302#define pte_young(pte) ((pte_val(pte) & _PAGE_A) != 0)
303#define pte_file(pte) ((pte_val(pte) & _PAGE_FILE) != 0)
304#define pte_special(pte) 0
305
306/*
307 * Note: we convert AR_RWX to AR_RX and AR_RW to AR_R by clearing the 2nd bit in the
308 * access rights:
309 */
310#define pte_wrprotect(pte) (__pte(pte_val(pte) & ~_PAGE_AR_RW))
311#define pte_mkwrite(pte) (__pte(pte_val(pte) | _PAGE_AR_RW))
312#define pte_mkold(pte) (__pte(pte_val(pte) & ~_PAGE_A))
313#define pte_mkyoung(pte) (__pte(pte_val(pte) | _PAGE_A))
314#define pte_mkclean(pte) (__pte(pte_val(pte) & ~_PAGE_D))
315#define pte_mkdirty(pte) (__pte(pte_val(pte) | _PAGE_D))
316#define pte_mkhuge(pte) (__pte(pte_val(pte)))
317#define pte_mkspecial(pte) (pte)
318
319/*
320 * Because ia64's Icache and Dcache is not coherent (on a cpu), we need to
321 * sync icache and dcache when we insert *new* executable page.
322 * __ia64_sync_icache_dcache() check Pg_arch_1 bit and flush icache
323 * if necessary.
324 *
325 * set_pte() is also called by the kernel, but we can expect that the kernel
326 * flushes icache explicitly if necessary.
327 */
328#define pte_present_exec_user(pte)\
329 ((pte_val(pte) & (_PAGE_P | _PAGE_PL_MASK | _PAGE_AR_RX)) == \
330 (_PAGE_P | _PAGE_PL_3 | _PAGE_AR_RX))
331
332extern void __ia64_sync_icache_dcache(pte_t pteval);
333static inline void set_pte(pte_t *ptep, pte_t pteval)
334{
335 /* page is present && page is user && page is executable
336 * && (page swapin or new page or page migraton
337 * || copy_on_write with page copying.)
338 */
339 if (pte_present_exec_user(pteval) &&
340 (!pte_present(*ptep) ||
341 pte_pfn(*ptep) != pte_pfn(pteval)))
342 /* load_module() calles flush_icache_range() explicitly*/
343 __ia64_sync_icache_dcache(pteval);
344 *ptep = pteval;
345}
346
347#define set_pte_at(mm,addr,ptep,pteval) set_pte(ptep,pteval)
348
349/*
350 * Make page protection values cacheable, uncacheable, or write-
351 * combining. Note that "protection" is really a misnomer here as the
352 * protection value contains the memory attribute bits, dirty bits, and
353 * various other bits as well.
354 */
355#define pgprot_cacheable(prot) __pgprot((pgprot_val(prot) & ~_PAGE_MA_MASK) | _PAGE_MA_WB)
356#define pgprot_noncached(prot) __pgprot((pgprot_val(prot) & ~_PAGE_MA_MASK) | _PAGE_MA_UC)
357#define pgprot_writecombine(prot) __pgprot((pgprot_val(prot) & ~_PAGE_MA_MASK) | _PAGE_MA_WC)
358
359struct file;
360extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
361 unsigned long size, pgprot_t vma_prot);
362#define __HAVE_PHYS_MEM_ACCESS_PROT
363
364static inline unsigned long
365pgd_index (unsigned long address)
366{
367 unsigned long region = address >> 61;
368 unsigned long l1index = (address >> PGDIR_SHIFT) & ((PTRS_PER_PGD >> 3) - 1);
369
370 return (region << (PAGE_SHIFT - 6)) | l1index;
371}
372
373/* The offset in the 1-level directory is given by the 3 region bits
374 (61..63) and the level-1 bits. */
375static inline pgd_t*
376pgd_offset (const struct mm_struct *mm, unsigned long address)
377{
378 return mm->pgd + pgd_index(address);
379}
380
381/* In the kernel's mapped region we completely ignore the region number
382 (since we know it's in region number 5). */
383#define pgd_offset_k(addr) \
384 (init_mm.pgd + (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1)))
385
386/* Look up a pgd entry in the gate area. On IA-64, the gate-area
387 resides in the kernel-mapped segment, hence we use pgd_offset_k()
388 here. */
389#define pgd_offset_gate(mm, addr) pgd_offset_k(addr)
390
391#ifdef CONFIG_PGTABLE_4
392/* Find an entry in the second-level page table.. */
393#define pud_offset(dir,addr) \
394 ((pud_t *) pgd_page_vaddr(*(dir)) + (((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1)))
395#endif
396
397/* Find an entry in the third-level page table.. */
398#define pmd_offset(dir,addr) \
399 ((pmd_t *) pud_page_vaddr(*(dir)) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
400
401/*
402 * Find an entry in the third-level page table. This looks more complicated than it
403 * should be because some platforms place page tables in high memory.
404 */
405#define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
406#define pte_offset_kernel(dir,addr) ((pte_t *) pmd_page_vaddr(*(dir)) + pte_index(addr))
407#define pte_offset_map(dir,addr) pte_offset_kernel(dir, addr)
408#define pte_unmap(pte) do { } while (0)
409
410/* atomic versions of the some PTE manipulations: */
411
412static inline int
413ptep_test_and_clear_young (struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
414{
415#ifdef CONFIG_SMP
416 if (!pte_young(*ptep))
417 return 0;
418 return test_and_clear_bit(_PAGE_A_BIT, ptep);
419#else
420 pte_t pte = *ptep;
421 if (!pte_young(pte))
422 return 0;
423 set_pte_at(vma->vm_mm, addr, ptep, pte_mkold(pte));
424 return 1;
425#endif
426}
427
428static inline pte_t
429ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
430{
431#ifdef CONFIG_SMP
432 return __pte(xchg((long *) ptep, 0));
433#else
434 pte_t pte = *ptep;
435 pte_clear(mm, addr, ptep);
436 return pte;
437#endif
438}
439
440static inline void
441ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
442{
443#ifdef CONFIG_SMP
444 unsigned long new, old;
445
446 do {
447 old = pte_val(*ptep);
448 new = pte_val(pte_wrprotect(__pte (old)));
449 } while (cmpxchg((unsigned long *) ptep, old, new) != old);
450#else
451 pte_t old_pte = *ptep;
452 set_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
453#endif
454}
455
456static inline int
457pte_same (pte_t a, pte_t b)
458{
459 return pte_val(a) == pte_val(b);
460}
461
462#define update_mmu_cache(vma, address, ptep) do { } while (0)
463
464extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
465extern void paging_init (void);
466
467/*
468 * Note: The macros below rely on the fact that MAX_SWAPFILES_SHIFT <= number of
469 * bits in the swap-type field of the swap pte. It would be nice to
470 * enforce that, but we can't easily include <linux/swap.h> here.
471 * (Of course, better still would be to define MAX_SWAPFILES_SHIFT here...).
472 *
473 * Format of swap pte:
474 * bit 0 : present bit (must be zero)
475 * bit 1 : _PAGE_FILE (must be zero)
476 * bits 2- 8: swap-type
477 * bits 9-62: swap offset
478 * bit 63 : _PAGE_PROTNONE bit
479 *
480 * Format of file pte:
481 * bit 0 : present bit (must be zero)
482 * bit 1 : _PAGE_FILE (must be one)
483 * bits 2-62: file_offset/PAGE_SIZE
484 * bit 63 : _PAGE_PROTNONE bit
485 */
486#define __swp_type(entry) (((entry).val >> 2) & 0x7f)
487#define __swp_offset(entry) (((entry).val << 1) >> 10)
488#define __swp_entry(type,offset) ((swp_entry_t) { ((type) << 2) | ((long) (offset) << 9) })
489#define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) })
490#define __swp_entry_to_pte(x) ((pte_t) { (x).val })
491
492#define PTE_FILE_MAX_BITS 61
493#define pte_to_pgoff(pte) ((pte_val(pte) << 1) >> 3)
494#define pgoff_to_pte(off) ((pte_t) { ((off) << 2) | _PAGE_FILE })
495
496/*
497 * ZERO_PAGE is a global shared page that is always zero: used
498 * for zero-mapped memory areas etc..
499 */
500extern unsigned long empty_zero_page[PAGE_SIZE/sizeof(unsigned long)];
501extern struct page *zero_page_memmap_ptr;
502#define ZERO_PAGE(vaddr) (zero_page_memmap_ptr)
503
504/* We provide our own get_unmapped_area to cope with VA holes for userland */
505#define HAVE_ARCH_UNMAPPED_AREA
506
507#ifdef CONFIG_HUGETLB_PAGE
508#define HUGETLB_PGDIR_SHIFT (HPAGE_SHIFT + 2*(PAGE_SHIFT-3))
509#define HUGETLB_PGDIR_SIZE (__IA64_UL(1) << HUGETLB_PGDIR_SHIFT)
510#define HUGETLB_PGDIR_MASK (~(HUGETLB_PGDIR_SIZE-1))
511#endif
512
513
514#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
515/*
516 * Update PTEP with ENTRY, which is guaranteed to be a less
517 * restrictive PTE. That is, ENTRY may have the ACCESSED, DIRTY, and
518 * WRITABLE bits turned on, when the value at PTEP did not. The
519 * WRITABLE bit may only be turned if SAFELY_WRITABLE is TRUE.
520 *
521 * SAFELY_WRITABLE is TRUE if we can update the value at PTEP without
522 * having to worry about races. On SMP machines, there are only two
523 * cases where this is true:
524 *
525 * (1) *PTEP has the PRESENT bit turned OFF
526 * (2) ENTRY has the DIRTY bit turned ON
527 *
528 * On ia64, we could implement this routine with a cmpxchg()-loop
529 * which ORs in the _PAGE_A/_PAGE_D bit if they're set in ENTRY.
530 * However, like on x86, we can get a more streamlined version by
531 * observing that it is OK to drop ACCESSED bit updates when
532 * SAFELY_WRITABLE is FALSE. Besides being rare, all that would do is
533 * result in an extra Access-bit fault, which would then turn on the
534 * ACCESSED bit in the low-level fault handler (iaccess_bit or
535 * daccess_bit in ivt.S).
536 */
537#ifdef CONFIG_SMP
538# define ptep_set_access_flags(__vma, __addr, __ptep, __entry, __safely_writable) \
539({ \
540 int __changed = !pte_same(*(__ptep), __entry); \
541 if (__changed && __safely_writable) { \
542 set_pte(__ptep, __entry); \
543 flush_tlb_page(__vma, __addr); \
544 } \
545 __changed; \
546})
547#else
548# define ptep_set_access_flags(__vma, __addr, __ptep, __entry, __safely_writable) \
549({ \
550 int __changed = !pte_same(*(__ptep), __entry); \
551 if (__changed) { \
552 set_pte_at((__vma)->vm_mm, (__addr), __ptep, __entry); \
553 flush_tlb_page(__vma, __addr); \
554 } \
555 __changed; \
556})
557#endif
558
559# ifdef CONFIG_VIRTUAL_MEM_MAP
560 /* arch mem_map init routine is needed due to holes in a virtual mem_map */
561# define __HAVE_ARCH_MEMMAP_INIT
562 extern void memmap_init (unsigned long size, int nid, unsigned long zone,
563 unsigned long start_pfn);
564# endif /* CONFIG_VIRTUAL_MEM_MAP */
565# endif /* !__ASSEMBLY__ */
566
567/*
568 * Identity-mapped regions use a large page size. We'll call such large pages
569 * "granules". If you can think of a better name that's unambiguous, let me
570 * know...
571 */
572#if defined(CONFIG_IA64_GRANULE_64MB)
573# define IA64_GRANULE_SHIFT _PAGE_SIZE_64M
574#elif defined(CONFIG_IA64_GRANULE_16MB)
575# define IA64_GRANULE_SHIFT _PAGE_SIZE_16M
576#endif
577#define IA64_GRANULE_SIZE (1 << IA64_GRANULE_SHIFT)
578/*
579 * log2() of the page size we use to map the kernel image (IA64_TR_KERNEL):
580 */
581#define KERNEL_TR_PAGE_SHIFT _PAGE_SIZE_64M
582#define KERNEL_TR_PAGE_SIZE (1 << KERNEL_TR_PAGE_SHIFT)
583
584/*
585 * No page table caches to initialise
586 */
587#define pgtable_cache_init() do { } while (0)
588
589/* These tell get_user_pages() that the first gate page is accessible from user-level. */
590#define FIXADDR_USER_START GATE_ADDR
591#ifdef HAVE_BUGGY_SEGREL
592# define FIXADDR_USER_END (GATE_ADDR + 2*PAGE_SIZE)
593#else
594# define FIXADDR_USER_END (GATE_ADDR + 2*PERCPU_PAGE_SIZE)
595#endif
596
597#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
598#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
599#define __HAVE_ARCH_PTEP_SET_WRPROTECT
600#define __HAVE_ARCH_PTE_SAME
601#define __HAVE_ARCH_PGD_OFFSET_GATE
602
603
604#ifndef CONFIG_PGTABLE_4
605#include <asm-generic/pgtable-nopud.h>
606#endif
607#include <asm-generic/pgtable.h>
608
609#endif /* _ASM_IA64_PGTABLE_H */
1#ifndef _ASM_IA64_PGTABLE_H
2#define _ASM_IA64_PGTABLE_H
3
4/*
5 * This file contains the functions and defines necessary to modify and use
6 * the IA-64 page table tree.
7 *
8 * This hopefully works with any (fixed) IA-64 page-size, as defined
9 * in <asm/page.h>.
10 *
11 * Copyright (C) 1998-2005 Hewlett-Packard Co
12 * David Mosberger-Tang <davidm@hpl.hp.com>
13 */
14
15
16#include <asm/mman.h>
17#include <asm/page.h>
18#include <asm/processor.h>
19#include <asm/types.h>
20
21#define IA64_MAX_PHYS_BITS 50 /* max. number of physical address bits (architected) */
22
23/*
24 * First, define the various bits in a PTE. Note that the PTE format
25 * matches the VHPT short format, the firt doubleword of the VHPD long
26 * format, and the first doubleword of the TLB insertion format.
27 */
28#define _PAGE_P_BIT 0
29#define _PAGE_A_BIT 5
30#define _PAGE_D_BIT 6
31
32#define _PAGE_P (1 << _PAGE_P_BIT) /* page present bit */
33#define _PAGE_MA_WB (0x0 << 2) /* write back memory attribute */
34#define _PAGE_MA_UC (0x4 << 2) /* uncacheable memory attribute */
35#define _PAGE_MA_UCE (0x5 << 2) /* UC exported attribute */
36#define _PAGE_MA_WC (0x6 << 2) /* write coalescing memory attribute */
37#define _PAGE_MA_NAT (0x7 << 2) /* not-a-thing attribute */
38#define _PAGE_MA_MASK (0x7 << 2)
39#define _PAGE_PL_0 (0 << 7) /* privilege level 0 (kernel) */
40#define _PAGE_PL_1 (1 << 7) /* privilege level 1 (unused) */
41#define _PAGE_PL_2 (2 << 7) /* privilege level 2 (unused) */
42#define _PAGE_PL_3 (3 << 7) /* privilege level 3 (user) */
43#define _PAGE_PL_MASK (3 << 7)
44#define _PAGE_AR_R (0 << 9) /* read only */
45#define _PAGE_AR_RX (1 << 9) /* read & execute */
46#define _PAGE_AR_RW (2 << 9) /* read & write */
47#define _PAGE_AR_RWX (3 << 9) /* read, write & execute */
48#define _PAGE_AR_R_RW (4 << 9) /* read / read & write */
49#define _PAGE_AR_RX_RWX (5 << 9) /* read & exec / read, write & exec */
50#define _PAGE_AR_RWX_RW (6 << 9) /* read, write & exec / read & write */
51#define _PAGE_AR_X_RX (7 << 9) /* exec & promote / read & exec */
52#define _PAGE_AR_MASK (7 << 9)
53#define _PAGE_AR_SHIFT 9
54#define _PAGE_A (1 << _PAGE_A_BIT) /* page accessed bit */
55#define _PAGE_D (1 << _PAGE_D_BIT) /* page dirty bit */
56#define _PAGE_PPN_MASK (((__IA64_UL(1) << IA64_MAX_PHYS_BITS) - 1) & ~0xfffUL)
57#define _PAGE_ED (__IA64_UL(1) << 52) /* exception deferral */
58#define _PAGE_PROTNONE (__IA64_UL(1) << 63)
59
60#define _PFN_MASK _PAGE_PPN_MASK
61/* Mask of bits which may be changed by pte_modify(); the odd bits are there for _PAGE_PROTNONE */
62#define _PAGE_CHG_MASK (_PAGE_P | _PAGE_PROTNONE | _PAGE_PL_MASK | _PAGE_AR_MASK | _PAGE_ED)
63
64#define _PAGE_SIZE_4K 12
65#define _PAGE_SIZE_8K 13
66#define _PAGE_SIZE_16K 14
67#define _PAGE_SIZE_64K 16
68#define _PAGE_SIZE_256K 18
69#define _PAGE_SIZE_1M 20
70#define _PAGE_SIZE_4M 22
71#define _PAGE_SIZE_16M 24
72#define _PAGE_SIZE_64M 26
73#define _PAGE_SIZE_256M 28
74#define _PAGE_SIZE_1G 30
75#define _PAGE_SIZE_4G 32
76
77#define __ACCESS_BITS _PAGE_ED | _PAGE_A | _PAGE_P | _PAGE_MA_WB
78#define __DIRTY_BITS_NO_ED _PAGE_A | _PAGE_P | _PAGE_D | _PAGE_MA_WB
79#define __DIRTY_BITS _PAGE_ED | __DIRTY_BITS_NO_ED
80
81/*
82 * How many pointers will a page table level hold expressed in shift
83 */
84#define PTRS_PER_PTD_SHIFT (PAGE_SHIFT-3)
85
86/*
87 * Definitions for fourth level:
88 */
89#define PTRS_PER_PTE (__IA64_UL(1) << (PTRS_PER_PTD_SHIFT))
90
91/*
92 * Definitions for third level:
93 *
94 * PMD_SHIFT determines the size of the area a third-level page table
95 * can map.
96 */
97#define PMD_SHIFT (PAGE_SHIFT + (PTRS_PER_PTD_SHIFT))
98#define PMD_SIZE (1UL << PMD_SHIFT)
99#define PMD_MASK (~(PMD_SIZE-1))
100#define PTRS_PER_PMD (1UL << (PTRS_PER_PTD_SHIFT))
101
102#if CONFIG_PGTABLE_LEVELS == 4
103/*
104 * Definitions for second level:
105 *
106 * PUD_SHIFT determines the size of the area a second-level page table
107 * can map.
108 */
109#define PUD_SHIFT (PMD_SHIFT + (PTRS_PER_PTD_SHIFT))
110#define PUD_SIZE (1UL << PUD_SHIFT)
111#define PUD_MASK (~(PUD_SIZE-1))
112#define PTRS_PER_PUD (1UL << (PTRS_PER_PTD_SHIFT))
113#endif
114
115/*
116 * Definitions for first level:
117 *
118 * PGDIR_SHIFT determines what a first-level page table entry can map.
119 */
120#if CONFIG_PGTABLE_LEVELS == 4
121#define PGDIR_SHIFT (PUD_SHIFT + (PTRS_PER_PTD_SHIFT))
122#else
123#define PGDIR_SHIFT (PMD_SHIFT + (PTRS_PER_PTD_SHIFT))
124#endif
125#define PGDIR_SIZE (__IA64_UL(1) << PGDIR_SHIFT)
126#define PGDIR_MASK (~(PGDIR_SIZE-1))
127#define PTRS_PER_PGD_SHIFT PTRS_PER_PTD_SHIFT
128#define PTRS_PER_PGD (1UL << PTRS_PER_PGD_SHIFT)
129#define USER_PTRS_PER_PGD (5*PTRS_PER_PGD/8) /* regions 0-4 are user regions */
130#define FIRST_USER_ADDRESS 0UL
131
132/*
133 * All the normal masks have the "page accessed" bits on, as any time
134 * they are used, the page is accessed. They are cleared only by the
135 * page-out routines.
136 */
137#define PAGE_NONE __pgprot(_PAGE_PROTNONE | _PAGE_A)
138#define PAGE_SHARED __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RW)
139#define PAGE_READONLY __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_R)
140#define PAGE_COPY __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_R)
141#define PAGE_COPY_EXEC __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RX)
142#define PAGE_GATE __pgprot(__ACCESS_BITS | _PAGE_PL_0 | _PAGE_AR_X_RX)
143#define PAGE_KERNEL __pgprot(__DIRTY_BITS | _PAGE_PL_0 | _PAGE_AR_RWX)
144#define PAGE_KERNELRX __pgprot(__ACCESS_BITS | _PAGE_PL_0 | _PAGE_AR_RX)
145#define PAGE_KERNEL_UC __pgprot(__DIRTY_BITS | _PAGE_PL_0 | _PAGE_AR_RWX | \
146 _PAGE_MA_UC)
147
148# ifndef __ASSEMBLY__
149
150#include <linux/sched.h> /* for mm_struct */
151#include <linux/bitops.h>
152#include <asm/cacheflush.h>
153#include <asm/mmu_context.h>
154
155/*
156 * Next come the mappings that determine how mmap() protection bits
157 * (PROT_EXEC, PROT_READ, PROT_WRITE, PROT_NONE) get implemented. The
158 * _P version gets used for a private shared memory segment, the _S
159 * version gets used for a shared memory segment with MAP_SHARED on.
160 * In a private shared memory segment, we do a copy-on-write if a task
161 * attempts to write to the page.
162 */
163 /* xwr */
164#define __P000 PAGE_NONE
165#define __P001 PAGE_READONLY
166#define __P010 PAGE_READONLY /* write to priv pg -> copy & make writable */
167#define __P011 PAGE_READONLY /* ditto */
168#define __P100 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_X_RX)
169#define __P101 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RX)
170#define __P110 PAGE_COPY_EXEC
171#define __P111 PAGE_COPY_EXEC
172
173#define __S000 PAGE_NONE
174#define __S001 PAGE_READONLY
175#define __S010 PAGE_SHARED /* we don't have (and don't need) write-only */
176#define __S011 PAGE_SHARED
177#define __S100 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_X_RX)
178#define __S101 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RX)
179#define __S110 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RWX)
180#define __S111 __pgprot(__ACCESS_BITS | _PAGE_PL_3 | _PAGE_AR_RWX)
181
182#define pgd_ERROR(e) printk("%s:%d: bad pgd %016lx.\n", __FILE__, __LINE__, pgd_val(e))
183#if CONFIG_PGTABLE_LEVELS == 4
184#define pud_ERROR(e) printk("%s:%d: bad pud %016lx.\n", __FILE__, __LINE__, pud_val(e))
185#endif
186#define pmd_ERROR(e) printk("%s:%d: bad pmd %016lx.\n", __FILE__, __LINE__, pmd_val(e))
187#define pte_ERROR(e) printk("%s:%d: bad pte %016lx.\n", __FILE__, __LINE__, pte_val(e))
188
189
190/*
191 * Some definitions to translate between mem_map, PTEs, and page addresses:
192 */
193
194
195/* Quick test to see if ADDR is a (potentially) valid physical address. */
196static inline long
197ia64_phys_addr_valid (unsigned long addr)
198{
199 return (addr & (local_cpu_data->unimpl_pa_mask)) == 0;
200}
201
202/*
203 * kern_addr_valid(ADDR) tests if ADDR is pointing to valid kernel
204 * memory. For the return value to be meaningful, ADDR must be >=
205 * PAGE_OFFSET. This operation can be relatively expensive (e.g.,
206 * require a hash-, or multi-level tree-lookup or something of that
207 * sort) but it guarantees to return TRUE only if accessing the page
208 * at that address does not cause an error. Note that there may be
209 * addresses for which kern_addr_valid() returns FALSE even though an
210 * access would not cause an error (e.g., this is typically true for
211 * memory mapped I/O regions.
212 *
213 * XXX Need to implement this for IA-64.
214 */
215#define kern_addr_valid(addr) (1)
216
217
218/*
219 * Now come the defines and routines to manage and access the three-level
220 * page table.
221 */
222
223
224#define VMALLOC_START (RGN_BASE(RGN_GATE) + 0x200000000UL)
225#ifdef CONFIG_VIRTUAL_MEM_MAP
226# define VMALLOC_END_INIT (RGN_BASE(RGN_GATE) + (1UL << (4*PAGE_SHIFT - 9)))
227extern unsigned long VMALLOC_END;
228#else
229#if defined(CONFIG_SPARSEMEM) && defined(CONFIG_SPARSEMEM_VMEMMAP)
230/* SPARSEMEM_VMEMMAP uses half of vmalloc... */
231# define VMALLOC_END (RGN_BASE(RGN_GATE) + (1UL << (4*PAGE_SHIFT - 10)))
232# define vmemmap ((struct page *)VMALLOC_END)
233#else
234# define VMALLOC_END (RGN_BASE(RGN_GATE) + (1UL << (4*PAGE_SHIFT - 9)))
235#endif
236#endif
237
238/* fs/proc/kcore.c */
239#define kc_vaddr_to_offset(v) ((v) - RGN_BASE(RGN_GATE))
240#define kc_offset_to_vaddr(o) ((o) + RGN_BASE(RGN_GATE))
241
242#define RGN_MAP_SHIFT (PGDIR_SHIFT + PTRS_PER_PGD_SHIFT - 3)
243#define RGN_MAP_LIMIT ((1UL << RGN_MAP_SHIFT) - PAGE_SIZE) /* per region addr limit */
244
245/*
246 * Conversion functions: convert page frame number (pfn) and a protection value to a page
247 * table entry (pte).
248 */
249#define pfn_pte(pfn, pgprot) \
250({ pte_t __pte; pte_val(__pte) = ((pfn) << PAGE_SHIFT) | pgprot_val(pgprot); __pte; })
251
252/* Extract pfn from pte. */
253#define pte_pfn(_pte) ((pte_val(_pte) & _PFN_MASK) >> PAGE_SHIFT)
254
255#define mk_pte(page, pgprot) pfn_pte(page_to_pfn(page), (pgprot))
256
257/* This takes a physical page address that is used by the remapping functions */
258#define mk_pte_phys(physpage, pgprot) \
259({ pte_t __pte; pte_val(__pte) = physpage + pgprot_val(pgprot); __pte; })
260
261#define pte_modify(_pte, newprot) \
262 (__pte((pte_val(_pte) & ~_PAGE_CHG_MASK) | (pgprot_val(newprot) & _PAGE_CHG_MASK)))
263
264#define pte_none(pte) (!pte_val(pte))
265#define pte_present(pte) (pte_val(pte) & (_PAGE_P | _PAGE_PROTNONE))
266#define pte_clear(mm,addr,pte) (pte_val(*(pte)) = 0UL)
267/* pte_page() returns the "struct page *" corresponding to the PTE: */
268#define pte_page(pte) virt_to_page(((pte_val(pte) & _PFN_MASK) + PAGE_OFFSET))
269
270#define pmd_none(pmd) (!pmd_val(pmd))
271#define pmd_bad(pmd) (!ia64_phys_addr_valid(pmd_val(pmd)))
272#define pmd_present(pmd) (pmd_val(pmd) != 0UL)
273#define pmd_clear(pmdp) (pmd_val(*(pmdp)) = 0UL)
274#define pmd_page_vaddr(pmd) ((unsigned long) __va(pmd_val(pmd) & _PFN_MASK))
275#define pmd_page(pmd) virt_to_page((pmd_val(pmd) + PAGE_OFFSET))
276
277#define pud_none(pud) (!pud_val(pud))
278#define pud_bad(pud) (!ia64_phys_addr_valid(pud_val(pud)))
279#define pud_present(pud) (pud_val(pud) != 0UL)
280#define pud_clear(pudp) (pud_val(*(pudp)) = 0UL)
281#define pud_page_vaddr(pud) ((unsigned long) __va(pud_val(pud) & _PFN_MASK))
282#define pud_page(pud) virt_to_page((pud_val(pud) + PAGE_OFFSET))
283
284#if CONFIG_PGTABLE_LEVELS == 4
285#define pgd_none(pgd) (!pgd_val(pgd))
286#define pgd_bad(pgd) (!ia64_phys_addr_valid(pgd_val(pgd)))
287#define pgd_present(pgd) (pgd_val(pgd) != 0UL)
288#define pgd_clear(pgdp) (pgd_val(*(pgdp)) = 0UL)
289#define pgd_page_vaddr(pgd) ((unsigned long) __va(pgd_val(pgd) & _PFN_MASK))
290#define pgd_page(pgd) virt_to_page((pgd_val(pgd) + PAGE_OFFSET))
291#endif
292
293/*
294 * The following have defined behavior only work if pte_present() is true.
295 */
296#define pte_write(pte) ((unsigned) (((pte_val(pte) & _PAGE_AR_MASK) >> _PAGE_AR_SHIFT) - 2) <= 4)
297#define pte_exec(pte) ((pte_val(pte) & _PAGE_AR_RX) != 0)
298#define pte_dirty(pte) ((pte_val(pte) & _PAGE_D) != 0)
299#define pte_young(pte) ((pte_val(pte) & _PAGE_A) != 0)
300#define pte_special(pte) 0
301
302/*
303 * Note: we convert AR_RWX to AR_RX and AR_RW to AR_R by clearing the 2nd bit in the
304 * access rights:
305 */
306#define pte_wrprotect(pte) (__pte(pte_val(pte) & ~_PAGE_AR_RW))
307#define pte_mkwrite(pte) (__pte(pte_val(pte) | _PAGE_AR_RW))
308#define pte_mkold(pte) (__pte(pte_val(pte) & ~_PAGE_A))
309#define pte_mkyoung(pte) (__pte(pte_val(pte) | _PAGE_A))
310#define pte_mkclean(pte) (__pte(pte_val(pte) & ~_PAGE_D))
311#define pte_mkdirty(pte) (__pte(pte_val(pte) | _PAGE_D))
312#define pte_mkhuge(pte) (__pte(pte_val(pte)))
313#define pte_mkspecial(pte) (pte)
314
315/*
316 * Because ia64's Icache and Dcache is not coherent (on a cpu), we need to
317 * sync icache and dcache when we insert *new* executable page.
318 * __ia64_sync_icache_dcache() check Pg_arch_1 bit and flush icache
319 * if necessary.
320 *
321 * set_pte() is also called by the kernel, but we can expect that the kernel
322 * flushes icache explicitly if necessary.
323 */
324#define pte_present_exec_user(pte)\
325 ((pte_val(pte) & (_PAGE_P | _PAGE_PL_MASK | _PAGE_AR_RX)) == \
326 (_PAGE_P | _PAGE_PL_3 | _PAGE_AR_RX))
327
328extern void __ia64_sync_icache_dcache(pte_t pteval);
329static inline void set_pte(pte_t *ptep, pte_t pteval)
330{
331 /* page is present && page is user && page is executable
332 * && (page swapin or new page or page migraton
333 * || copy_on_write with page copying.)
334 */
335 if (pte_present_exec_user(pteval) &&
336 (!pte_present(*ptep) ||
337 pte_pfn(*ptep) != pte_pfn(pteval)))
338 /* load_module() calles flush_icache_range() explicitly*/
339 __ia64_sync_icache_dcache(pteval);
340 *ptep = pteval;
341}
342
343#define set_pte_at(mm,addr,ptep,pteval) set_pte(ptep,pteval)
344
345/*
346 * Make page protection values cacheable, uncacheable, or write-
347 * combining. Note that "protection" is really a misnomer here as the
348 * protection value contains the memory attribute bits, dirty bits, and
349 * various other bits as well.
350 */
351#define pgprot_cacheable(prot) __pgprot((pgprot_val(prot) & ~_PAGE_MA_MASK) | _PAGE_MA_WB)
352#define pgprot_noncached(prot) __pgprot((pgprot_val(prot) & ~_PAGE_MA_MASK) | _PAGE_MA_UC)
353#define pgprot_writecombine(prot) __pgprot((pgprot_val(prot) & ~_PAGE_MA_MASK) | _PAGE_MA_WC)
354
355struct file;
356extern pgprot_t phys_mem_access_prot(struct file *file, unsigned long pfn,
357 unsigned long size, pgprot_t vma_prot);
358#define __HAVE_PHYS_MEM_ACCESS_PROT
359
360static inline unsigned long
361pgd_index (unsigned long address)
362{
363 unsigned long region = address >> 61;
364 unsigned long l1index = (address >> PGDIR_SHIFT) & ((PTRS_PER_PGD >> 3) - 1);
365
366 return (region << (PAGE_SHIFT - 6)) | l1index;
367}
368
369/* The offset in the 1-level directory is given by the 3 region bits
370 (61..63) and the level-1 bits. */
371static inline pgd_t*
372pgd_offset (const struct mm_struct *mm, unsigned long address)
373{
374 return mm->pgd + pgd_index(address);
375}
376
377/* In the kernel's mapped region we completely ignore the region number
378 (since we know it's in region number 5). */
379#define pgd_offset_k(addr) \
380 (init_mm.pgd + (((addr) >> PGDIR_SHIFT) & (PTRS_PER_PGD - 1)))
381
382/* Look up a pgd entry in the gate area. On IA-64, the gate-area
383 resides in the kernel-mapped segment, hence we use pgd_offset_k()
384 here. */
385#define pgd_offset_gate(mm, addr) pgd_offset_k(addr)
386
387#if CONFIG_PGTABLE_LEVELS == 4
388/* Find an entry in the second-level page table.. */
389#define pud_offset(dir,addr) \
390 ((pud_t *) pgd_page_vaddr(*(dir)) + (((addr) >> PUD_SHIFT) & (PTRS_PER_PUD - 1)))
391#endif
392
393/* Find an entry in the third-level page table.. */
394#define pmd_offset(dir,addr) \
395 ((pmd_t *) pud_page_vaddr(*(dir)) + (((addr) >> PMD_SHIFT) & (PTRS_PER_PMD - 1)))
396
397/*
398 * Find an entry in the third-level page table. This looks more complicated than it
399 * should be because some platforms place page tables in high memory.
400 */
401#define pte_index(addr) (((addr) >> PAGE_SHIFT) & (PTRS_PER_PTE - 1))
402#define pte_offset_kernel(dir,addr) ((pte_t *) pmd_page_vaddr(*(dir)) + pte_index(addr))
403#define pte_offset_map(dir,addr) pte_offset_kernel(dir, addr)
404#define pte_unmap(pte) do { } while (0)
405
406/* atomic versions of the some PTE manipulations: */
407
408static inline int
409ptep_test_and_clear_young (struct vm_area_struct *vma, unsigned long addr, pte_t *ptep)
410{
411#ifdef CONFIG_SMP
412 if (!pte_young(*ptep))
413 return 0;
414 return test_and_clear_bit(_PAGE_A_BIT, ptep);
415#else
416 pte_t pte = *ptep;
417 if (!pte_young(pte))
418 return 0;
419 set_pte_at(vma->vm_mm, addr, ptep, pte_mkold(pte));
420 return 1;
421#endif
422}
423
424static inline pte_t
425ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
426{
427#ifdef CONFIG_SMP
428 return __pte(xchg((long *) ptep, 0));
429#else
430 pte_t pte = *ptep;
431 pte_clear(mm, addr, ptep);
432 return pte;
433#endif
434}
435
436static inline void
437ptep_set_wrprotect(struct mm_struct *mm, unsigned long addr, pte_t *ptep)
438{
439#ifdef CONFIG_SMP
440 unsigned long new, old;
441
442 do {
443 old = pte_val(*ptep);
444 new = pte_val(pte_wrprotect(__pte (old)));
445 } while (cmpxchg((unsigned long *) ptep, old, new) != old);
446#else
447 pte_t old_pte = *ptep;
448 set_pte_at(mm, addr, ptep, pte_wrprotect(old_pte));
449#endif
450}
451
452static inline int
453pte_same (pte_t a, pte_t b)
454{
455 return pte_val(a) == pte_val(b);
456}
457
458#define update_mmu_cache(vma, address, ptep) do { } while (0)
459
460extern pgd_t swapper_pg_dir[PTRS_PER_PGD];
461extern void paging_init (void);
462
463/*
464 * Note: The macros below rely on the fact that MAX_SWAPFILES_SHIFT <= number of
465 * bits in the swap-type field of the swap pte. It would be nice to
466 * enforce that, but we can't easily include <linux/swap.h> here.
467 * (Of course, better still would be to define MAX_SWAPFILES_SHIFT here...).
468 *
469 * Format of swap pte:
470 * bit 0 : present bit (must be zero)
471 * bits 1- 7: swap-type
472 * bits 8-62: swap offset
473 * bit 63 : _PAGE_PROTNONE bit
474 */
475#define __swp_type(entry) (((entry).val >> 1) & 0x7f)
476#define __swp_offset(entry) (((entry).val << 1) >> 9)
477#define __swp_entry(type,offset) ((swp_entry_t) { ((type) << 1) | ((long) (offset) << 8) })
478#define __pte_to_swp_entry(pte) ((swp_entry_t) { pte_val(pte) })
479#define __swp_entry_to_pte(x) ((pte_t) { (x).val })
480
481/*
482 * ZERO_PAGE is a global shared page that is always zero: used
483 * for zero-mapped memory areas etc..
484 */
485extern unsigned long empty_zero_page[PAGE_SIZE/sizeof(unsigned long)];
486extern struct page *zero_page_memmap_ptr;
487#define ZERO_PAGE(vaddr) (zero_page_memmap_ptr)
488
489/* We provide our own get_unmapped_area to cope with VA holes for userland */
490#define HAVE_ARCH_UNMAPPED_AREA
491
492#ifdef CONFIG_HUGETLB_PAGE
493#define HUGETLB_PGDIR_SHIFT (HPAGE_SHIFT + 2*(PAGE_SHIFT-3))
494#define HUGETLB_PGDIR_SIZE (__IA64_UL(1) << HUGETLB_PGDIR_SHIFT)
495#define HUGETLB_PGDIR_MASK (~(HUGETLB_PGDIR_SIZE-1))
496#endif
497
498
499#define __HAVE_ARCH_PTEP_SET_ACCESS_FLAGS
500/*
501 * Update PTEP with ENTRY, which is guaranteed to be a less
502 * restrictive PTE. That is, ENTRY may have the ACCESSED, DIRTY, and
503 * WRITABLE bits turned on, when the value at PTEP did not. The
504 * WRITABLE bit may only be turned if SAFELY_WRITABLE is TRUE.
505 *
506 * SAFELY_WRITABLE is TRUE if we can update the value at PTEP without
507 * having to worry about races. On SMP machines, there are only two
508 * cases where this is true:
509 *
510 * (1) *PTEP has the PRESENT bit turned OFF
511 * (2) ENTRY has the DIRTY bit turned ON
512 *
513 * On ia64, we could implement this routine with a cmpxchg()-loop
514 * which ORs in the _PAGE_A/_PAGE_D bit if they're set in ENTRY.
515 * However, like on x86, we can get a more streamlined version by
516 * observing that it is OK to drop ACCESSED bit updates when
517 * SAFELY_WRITABLE is FALSE. Besides being rare, all that would do is
518 * result in an extra Access-bit fault, which would then turn on the
519 * ACCESSED bit in the low-level fault handler (iaccess_bit or
520 * daccess_bit in ivt.S).
521 */
522#ifdef CONFIG_SMP
523# define ptep_set_access_flags(__vma, __addr, __ptep, __entry, __safely_writable) \
524({ \
525 int __changed = !pte_same(*(__ptep), __entry); \
526 if (__changed && __safely_writable) { \
527 set_pte(__ptep, __entry); \
528 flush_tlb_page(__vma, __addr); \
529 } \
530 __changed; \
531})
532#else
533# define ptep_set_access_flags(__vma, __addr, __ptep, __entry, __safely_writable) \
534({ \
535 int __changed = !pte_same(*(__ptep), __entry); \
536 if (__changed) { \
537 set_pte_at((__vma)->vm_mm, (__addr), __ptep, __entry); \
538 flush_tlb_page(__vma, __addr); \
539 } \
540 __changed; \
541})
542#endif
543
544# ifdef CONFIG_VIRTUAL_MEM_MAP
545 /* arch mem_map init routine is needed due to holes in a virtual mem_map */
546# define __HAVE_ARCH_MEMMAP_INIT
547 extern void memmap_init (unsigned long size, int nid, unsigned long zone,
548 unsigned long start_pfn);
549# endif /* CONFIG_VIRTUAL_MEM_MAP */
550# endif /* !__ASSEMBLY__ */
551
552/*
553 * Identity-mapped regions use a large page size. We'll call such large pages
554 * "granules". If you can think of a better name that's unambiguous, let me
555 * know...
556 */
557#if defined(CONFIG_IA64_GRANULE_64MB)
558# define IA64_GRANULE_SHIFT _PAGE_SIZE_64M
559#elif defined(CONFIG_IA64_GRANULE_16MB)
560# define IA64_GRANULE_SHIFT _PAGE_SIZE_16M
561#endif
562#define IA64_GRANULE_SIZE (1 << IA64_GRANULE_SHIFT)
563/*
564 * log2() of the page size we use to map the kernel image (IA64_TR_KERNEL):
565 */
566#define KERNEL_TR_PAGE_SHIFT _PAGE_SIZE_64M
567#define KERNEL_TR_PAGE_SIZE (1 << KERNEL_TR_PAGE_SHIFT)
568
569/*
570 * No page table caches to initialise
571 */
572#define pgtable_cache_init() do { } while (0)
573
574/* These tell get_user_pages() that the first gate page is accessible from user-level. */
575#define FIXADDR_USER_START GATE_ADDR
576#ifdef HAVE_BUGGY_SEGREL
577# define FIXADDR_USER_END (GATE_ADDR + 2*PAGE_SIZE)
578#else
579# define FIXADDR_USER_END (GATE_ADDR + 2*PERCPU_PAGE_SIZE)
580#endif
581
582#define __HAVE_ARCH_PTEP_TEST_AND_CLEAR_YOUNG
583#define __HAVE_ARCH_PTEP_GET_AND_CLEAR
584#define __HAVE_ARCH_PTEP_SET_WRPROTECT
585#define __HAVE_ARCH_PTE_SAME
586#define __HAVE_ARCH_PGD_OFFSET_GATE
587
588
589#if CONFIG_PGTABLE_LEVELS == 3
590#include <asm-generic/pgtable-nopud.h>
591#endif
592#include <asm-generic/pgtable.h>
593
594#endif /* _ASM_IA64_PGTABLE_H */