2 * This file contains common routines for dealing with free of page tables
3 * Along with common page table handling code
5 * Derived from arch/powerpc/mm/tlb_64.c:
6 * Copyright (C) 1995-1996 Gary Thomas (gdt@linuxppc.org)
8 * Modifications by Paul Mackerras (PowerMac) (paulus@cs.anu.edu.au)
9 * and Cort Dougan (PReP) (cort@cs.nmt.edu)
10 * Copyright (C) 1996 Paul Mackerras
12 * Derived from "arch/i386/mm/init.c"
13 * Copyright (C) 1991, 1992, 1993, 1994 Linus Torvalds
15 * Dave Engebretsen <engebret@us.ibm.com>
16 * Rework for PPC64 port.
18 * This program is free software; you can redistribute it and/or
19 * modify it under the terms of the GNU General Public License
20 * as published by the Free Software Foundation; either version
21 * 2 of the License, or (at your option) any later version.
24 #include <linux/kernel.h>
25 #include <linux/gfp.h>
27 #include <linux/init.h>
28 #include <linux/percpu.h>
29 #include <linux/hardirq.h>
30 #include <linux/hugetlb.h>
31 #include <asm/pgalloc.h>
32 #include <asm/tlbflush.h>
37 static inline int is_exec_fault(void)
39 return current
->thread
.regs
&& TRAP(current
->thread
.regs
) == 0x400;
42 /* We only try to do i/d cache coherency on stuff that looks like
43 * reasonably "normal" PTEs. We currently require a PTE to be present
44 * and we avoid _PAGE_SPECIAL and _PAGE_NO_CACHE. We also only do that
47 static inline int pte_looks_normal(pte_t pte
)
49 return (pte_val(pte
) &
50 (_PAGE_PRESENT
| _PAGE_SPECIAL
| _PAGE_NO_CACHE
| _PAGE_USER
)) ==
51 (_PAGE_PRESENT
| _PAGE_USER
);
54 struct page
* maybe_pte_to_page(pte_t pte
)
56 unsigned long pfn
= pte_pfn(pte
);
59 if (unlikely(!pfn_valid(pfn
)))
61 page
= pfn_to_page(pfn
);
62 if (PageReserved(page
))
67 #if defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0
69 /* Server-style MMU handles coherency when hashing if HW exec permission
70 * is supposed per page (currently 64-bit only). If not, then, we always
71 * flush the cache for valid PTEs in set_pte. Embedded CPU without HW exec
72 * support falls into the same category.
75 static pte_t
set_pte_filter(pte_t pte
, unsigned long addr
)
77 pte
= __pte(pte_val(pte
) & ~_PAGE_HPTEFLAGS
);
78 if (pte_looks_normal(pte
) && !(cpu_has_feature(CPU_FTR_COHERENT_ICACHE
) ||
79 cpu_has_feature(CPU_FTR_NOEXECUTE
))) {
80 struct page
*pg
= maybe_pte_to_page(pte
);
83 if (!test_bit(PG_arch_1
, &pg
->flags
)) {
85 /* On 8xx, cache control instructions (particularly
86 * "dcbst" from flush_dcache_icache) fault as write
87 * operation if there is an unpopulated TLB entry
88 * for the address in question. To workaround that,
89 * we invalidate the TLB here, thus avoiding dcbst
92 /* 8xx doesn't care about PID, size or ind args */
93 _tlbil_va(addr
, 0, 0, 0);
94 #endif /* CONFIG_8xx */
95 flush_dcache_icache_page(pg
);
96 set_bit(PG_arch_1
, &pg
->flags
);
102 static pte_t
set_access_flags_filter(pte_t pte
, struct vm_area_struct
*vma
,
108 #else /* defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0 */
110 /* Embedded type MMU with HW exec support. This is a bit more complicated
111 * as we don't have two bits to spare for _PAGE_EXEC and _PAGE_HWEXEC so
112 * instead we "filter out" the exec permission for non clean pages.
114 static pte_t
set_pte_filter(pte_t pte
, unsigned long addr
)
118 /* No exec permission in the first place, move on */
119 if (!(pte_val(pte
) & _PAGE_EXEC
) || !pte_looks_normal(pte
))
122 /* If you set _PAGE_EXEC on weird pages you're on your own */
123 pg
= maybe_pte_to_page(pte
);
127 /* If the page clean, we move on */
128 if (test_bit(PG_arch_1
, &pg
->flags
))
131 /* If it's an exec fault, we flush the cache and make it clean */
132 if (is_exec_fault()) {
133 flush_dcache_icache_page(pg
);
134 set_bit(PG_arch_1
, &pg
->flags
);
138 /* Else, we filter out _PAGE_EXEC */
139 return __pte(pte_val(pte
) & ~_PAGE_EXEC
);
142 static pte_t
set_access_flags_filter(pte_t pte
, struct vm_area_struct
*vma
,
147 /* So here, we only care about exec faults, as we use them
148 * to recover lost _PAGE_EXEC and perform I$/D$ coherency
149 * if necessary. Also if _PAGE_EXEC is already set, same deal,
152 if (dirty
|| (pte_val(pte
) & _PAGE_EXEC
) || !is_exec_fault())
155 #ifdef CONFIG_DEBUG_VM
156 /* So this is an exec fault, _PAGE_EXEC is not set. If it was
157 * an error we would have bailed out earlier in do_page_fault()
158 * but let's make sure of it
160 if (WARN_ON(!(vma
->vm_flags
& VM_EXEC
)))
162 #endif /* CONFIG_DEBUG_VM */
164 /* If you set _PAGE_EXEC on weird pages you're on your own */
165 pg
= maybe_pte_to_page(pte
);
169 /* If the page is already clean, we move on */
170 if (test_bit(PG_arch_1
, &pg
->flags
))
173 /* Clean the page and set PG_arch_1 */
174 flush_dcache_icache_page(pg
);
175 set_bit(PG_arch_1
, &pg
->flags
);
178 return __pte(pte_val(pte
) | _PAGE_EXEC
);
181 #endif /* !(defined(CONFIG_PPC_STD_MMU) || _PAGE_EXEC == 0) */
184 * set_pte stores a linux PTE into the linux page table.
186 void set_pte_at(struct mm_struct
*mm
, unsigned long addr
, pte_t
*ptep
,
189 #ifdef CONFIG_DEBUG_VM
190 WARN_ON(pte_present(*ptep
));
192 /* Note: mm->context.id might not yet have been assigned as
193 * this context might not have been activated yet when this
196 pte
= set_pte_filter(pte
, addr
);
198 /* Perform the setting of the PTE */
199 __set_pte_at(mm
, addr
, ptep
, pte
, 0);
203 * This is called when relaxing access to a PTE. It's also called in the page
204 * fault path when we don't hit any of the major fault cases, ie, a minor
205 * update of _PAGE_ACCESSED, _PAGE_DIRTY, etc... The generic code will have
206 * handled those two for us, we additionally deal with missing execute
207 * permission here on some processors
209 int ptep_set_access_flags(struct vm_area_struct
*vma
, unsigned long address
,
210 pte_t
*ptep
, pte_t entry
, int dirty
)
213 entry
= set_access_flags_filter(entry
, vma
, dirty
);
214 changed
= !pte_same(*(ptep
), entry
);
216 if (!is_vm_hugetlb_page(vma
))
217 assert_pte_locked(vma
->vm_mm
, address
);
218 __ptep_set_access_flags(ptep
, entry
);
219 flush_tlb_page_nohash(vma
, address
);
224 #ifdef CONFIG_DEBUG_VM
225 void assert_pte_locked(struct mm_struct
*mm
, unsigned long addr
)
233 pgd
= mm
->pgd
+ pgd_index(addr
);
234 BUG_ON(pgd_none(*pgd
));
235 pud
= pud_offset(pgd
, addr
);
236 BUG_ON(pud_none(*pud
));
237 pmd
= pmd_offset(pud
, addr
);
239 * khugepaged to collapse normal pages to hugepage, first set
240 * pmd to none to force page fault/gup to take mmap_sem. After
241 * pmd is set to none, we do a pte_clear which does this assertion
242 * so if we find pmd none, return.
246 BUG_ON(!pmd_present(*pmd
));
247 assert_spin_locked(pte_lockptr(mm
, pmd
));
249 #endif /* CONFIG_DEBUG_VM */