3 #include <asm/pgalloc.h>
4 #include <asm/pgtable.h>
6 #include <asm/fixmap.h>
8 #define PGALLOC_GFP GFP_KERNEL | __GFP_NOTRACK | __GFP_REPEAT | __GFP_ZERO
11 #define PGALLOC_USER_GFP __GFP_HIGHMEM
13 #define PGALLOC_USER_GFP 0
16 gfp_t __userpte_alloc_gfp
= PGALLOC_GFP
| PGALLOC_USER_GFP
;
18 pte_t
*pte_alloc_one_kernel(struct mm_struct
*mm
, unsigned long address
)
20 return (pte_t
*)__get_free_page(PGALLOC_GFP
);
23 pgtable_t
pte_alloc_one(struct mm_struct
*mm
, unsigned long address
)
27 pte
= alloc_pages(__userpte_alloc_gfp
, 0);
29 pgtable_page_ctor(pte
);
33 static int __init
setup_userpte(char *arg
)
39 * "userpte=nohigh" disables allocation of user pagetables in
42 if (strcmp(arg
, "nohigh") == 0)
43 __userpte_alloc_gfp
&= ~__GFP_HIGHMEM
;
48 early_param("userpte", setup_userpte
);
50 void ___pte_free_tlb(struct mmu_gather
*tlb
, struct page
*pte
)
52 pgtable_page_dtor(pte
);
53 paravirt_release_pte(page_to_pfn(pte
));
54 tlb_remove_page(tlb
, pte
);
57 #if PAGETABLE_LEVELS > 2
58 void ___pmd_free_tlb(struct mmu_gather
*tlb
, pmd_t
*pmd
)
60 paravirt_release_pmd(__pa(pmd
) >> PAGE_SHIFT
);
62 * NOTE! For PAE, any changes to the top page-directory-pointer-table
63 * entries need a full cr3 reload to flush.
66 tlb
->need_flush_all
= 1;
68 tlb_remove_page(tlb
, virt_to_page(pmd
));
71 #if PAGETABLE_LEVELS > 3
72 void ___pud_free_tlb(struct mmu_gather
*tlb
, pud_t
*pud
)
74 paravirt_release_pud(__pa(pud
) >> PAGE_SHIFT
);
75 tlb_remove_page(tlb
, virt_to_page(pud
));
77 #endif /* PAGETABLE_LEVELS > 3 */
78 #endif /* PAGETABLE_LEVELS > 2 */
80 static inline void pgd_list_add(pgd_t
*pgd
)
82 struct page
*page
= virt_to_page(pgd
);
84 list_add(&page
->lru
, &pgd_list
);
87 static inline void pgd_list_del(pgd_t
*pgd
)
89 struct page
*page
= virt_to_page(pgd
);
94 #define UNSHARED_PTRS_PER_PGD \
95 (SHARED_KERNEL_PMD ? KERNEL_PGD_BOUNDARY : PTRS_PER_PGD)
98 static void pgd_set_mm(pgd_t
*pgd
, struct mm_struct
*mm
)
100 BUILD_BUG_ON(sizeof(virt_to_page(pgd
)->index
) < sizeof(mm
));
101 virt_to_page(pgd
)->index
= (pgoff_t
)mm
;
104 struct mm_struct
*pgd_page_get_mm(struct page
*page
)
106 return (struct mm_struct
*)page
->index
;
109 static void pgd_ctor(struct mm_struct
*mm
, pgd_t
*pgd
)
111 /* If the pgd points to a shared pagetable level (either the
112 ptes in non-PAE, or shared PMD in PAE), then just copy the
113 references from swapper_pg_dir. */
114 if (PAGETABLE_LEVELS
== 2 ||
115 (PAGETABLE_LEVELS
== 3 && SHARED_KERNEL_PMD
) ||
116 PAGETABLE_LEVELS
== 4) {
117 clone_pgd_range(pgd
+ KERNEL_PGD_BOUNDARY
,
118 swapper_pg_dir
+ KERNEL_PGD_BOUNDARY
,
122 /* list required to sync kernel mapping updates */
123 if (!SHARED_KERNEL_PMD
) {
129 static void pgd_dtor(pgd_t
*pgd
)
131 if (SHARED_KERNEL_PMD
)
134 spin_lock(&pgd_lock
);
136 spin_unlock(&pgd_lock
);
140 * List of all pgd's needed for non-PAE so it can invalidate entries
141 * in both cached and uncached pgd's; not needed for PAE since the
142 * kernel pmd is shared. If PAE were not to share the pmd a similar
143 * tactic would be needed. This is essentially codepath-based locking
144 * against pageattr.c; it is the unique case in which a valid change
145 * of kernel pagetables can't be lazily synchronized by vmalloc faults.
146 * vmalloc faults work because attached pagetables are never freed.
150 #ifdef CONFIG_X86_PAE
152 * In PAE mode, we need to do a cr3 reload (=tlb flush) when
153 * updating the top-level pagetable entries to guarantee the
154 * processor notices the update. Since this is expensive, and
155 * all 4 top-level entries are used almost immediately in a
156 * new process's life, we just pre-populate them here.
158 * Also, if we're in a paravirt environment where the kernel pmd is
159 * not shared between pagetables (!SHARED_KERNEL_PMDS), we allocate
160 * and initialize the kernel pmds here.
162 #define PREALLOCATED_PMDS UNSHARED_PTRS_PER_PGD
164 void pud_populate(struct mm_struct
*mm
, pud_t
*pudp
, pmd_t
*pmd
)
166 paravirt_alloc_pmd(mm
, __pa(pmd
) >> PAGE_SHIFT
);
168 /* Note: almost everything apart from _PAGE_PRESENT is
169 reserved at the pmd (PDPT) level. */
170 set_pud(pudp
, __pud(__pa(pmd
) | _PAGE_PRESENT
));
173 * According to Intel App note "TLBs, Paging-Structure Caches,
174 * and Their Invalidation", April 2007, document 317080-001,
175 * section 8.1: in PAE mode we explicitly have to flush the
176 * TLB via cr3 if the top-level pgd is changed...
180 #else /* !CONFIG_X86_PAE */
182 /* No need to prepopulate any pagetable entries in non-PAE modes. */
183 #define PREALLOCATED_PMDS 0
185 #endif /* CONFIG_X86_PAE */
187 static void free_pmds(pmd_t
*pmds
[])
191 for(i
= 0; i
< PREALLOCATED_PMDS
; i
++)
193 pgtable_pmd_page_dtor(virt_to_page(pmds
[i
]));
194 free_page((unsigned long)pmds
[i
]);
198 static int preallocate_pmds(pmd_t
*pmds
[])
203 for(i
= 0; i
< PREALLOCATED_PMDS
; i
++) {
204 pmd_t
*pmd
= (pmd_t
*)__get_free_page(PGALLOC_GFP
);
207 if (pmd
&& !pgtable_pmd_page_ctor(virt_to_page(pmd
))) {
208 free_page((unsigned long)pmds
[i
]);
224 * Mop up any pmd pages which may still be attached to the pgd.
225 * Normally they will be freed by munmap/exit_mmap, but any pmd we
226 * preallocate which never got a corresponding vma will need to be
229 static void pgd_mop_up_pmds(struct mm_struct
*mm
, pgd_t
*pgdp
)
233 for(i
= 0; i
< PREALLOCATED_PMDS
; i
++) {
236 if (pgd_val(pgd
) != 0) {
237 pmd_t
*pmd
= (pmd_t
*)pgd_page_vaddr(pgd
);
239 pgdp
[i
] = native_make_pgd(0);
241 paravirt_release_pmd(pgd_val(pgd
) >> PAGE_SHIFT
);
247 static void pgd_prepopulate_pmd(struct mm_struct
*mm
, pgd_t
*pgd
, pmd_t
*pmds
[])
252 if (PREALLOCATED_PMDS
== 0) /* Work around gcc-3.4.x bug */
255 pud
= pud_offset(pgd
, 0);
257 for (i
= 0; i
< PREALLOCATED_PMDS
; i
++, pud
++) {
258 pmd_t
*pmd
= pmds
[i
];
260 if (i
>= KERNEL_PGD_BOUNDARY
)
261 memcpy(pmd
, (pmd_t
*)pgd_page_vaddr(swapper_pg_dir
[i
]),
262 sizeof(pmd_t
) * PTRS_PER_PMD
);
264 pud_populate(mm
, pud
, pmd
);
268 pgd_t
*pgd_alloc(struct mm_struct
*mm
)
271 pmd_t
*pmds
[PREALLOCATED_PMDS
];
273 pgd
= (pgd_t
*)__get_free_page(PGALLOC_GFP
);
280 if (preallocate_pmds(pmds
) != 0)
283 if (paravirt_pgd_alloc(mm
) != 0)
287 * Make sure that pre-populating the pmds is atomic with
288 * respect to anything walking the pgd_list, so that they
289 * never see a partially populated pgd.
291 spin_lock(&pgd_lock
);
294 pgd_prepopulate_pmd(mm
, pgd
, pmds
);
296 spin_unlock(&pgd_lock
);
303 free_page((unsigned long)pgd
);
308 void pgd_free(struct mm_struct
*mm
, pgd_t
*pgd
)
310 pgd_mop_up_pmds(mm
, pgd
);
312 paravirt_pgd_free(mm
, pgd
);
313 free_page((unsigned long)pgd
);
317 * Used to set accessed or dirty bits in the page table entries
318 * on other architectures. On x86, the accessed and dirty bits
319 * are tracked by hardware. However, do_wp_page calls this function
320 * to also make the pte writeable at the same time the dirty bit is
321 * set. In that case we do actually need to write the PTE.
323 int ptep_set_access_flags(struct vm_area_struct
*vma
,
324 unsigned long address
, pte_t
*ptep
,
325 pte_t entry
, int dirty
)
327 int changed
= !pte_same(*ptep
, entry
);
329 if (changed
&& dirty
) {
331 pte_update_defer(vma
->vm_mm
, address
, ptep
);
337 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
338 int pmdp_set_access_flags(struct vm_area_struct
*vma
,
339 unsigned long address
, pmd_t
*pmdp
,
340 pmd_t entry
, int dirty
)
342 int changed
= !pmd_same(*pmdp
, entry
);
344 VM_BUG_ON(address
& ~HPAGE_PMD_MASK
);
346 if (changed
&& dirty
) {
348 pmd_update_defer(vma
->vm_mm
, address
, pmdp
);
350 * We had a write-protection fault here and changed the pmd
351 * to to more permissive. No need to flush the TLB for that,
352 * #PF is architecturally guaranteed to do that and in the
353 * worst-case we'll generate a spurious fault.
361 int ptep_test_and_clear_young(struct vm_area_struct
*vma
,
362 unsigned long addr
, pte_t
*ptep
)
366 if (pte_young(*ptep
))
367 ret
= test_and_clear_bit(_PAGE_BIT_ACCESSED
,
368 (unsigned long *) &ptep
->pte
);
371 pte_update(vma
->vm_mm
, addr
, ptep
);
376 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
377 int pmdp_test_and_clear_young(struct vm_area_struct
*vma
,
378 unsigned long addr
, pmd_t
*pmdp
)
382 if (pmd_young(*pmdp
))
383 ret
= test_and_clear_bit(_PAGE_BIT_ACCESSED
,
384 (unsigned long *)pmdp
);
387 pmd_update(vma
->vm_mm
, addr
, pmdp
);
393 int ptep_clear_flush_young(struct vm_area_struct
*vma
,
394 unsigned long address
, pte_t
*ptep
)
398 young
= ptep_test_and_clear_young(vma
, address
, ptep
);
400 flush_tlb_page(vma
, address
);
405 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
406 int pmdp_clear_flush_young(struct vm_area_struct
*vma
,
407 unsigned long address
, pmd_t
*pmdp
)
411 VM_BUG_ON(address
& ~HPAGE_PMD_MASK
);
413 young
= pmdp_test_and_clear_young(vma
, address
, pmdp
);
415 flush_tlb_range(vma
, address
, address
+ HPAGE_PMD_SIZE
);
420 void pmdp_splitting_flush(struct vm_area_struct
*vma
,
421 unsigned long address
, pmd_t
*pmdp
)
424 VM_BUG_ON(address
& ~HPAGE_PMD_MASK
);
425 set
= !test_and_set_bit(_PAGE_BIT_SPLITTING
,
426 (unsigned long *)pmdp
);
428 pmd_update(vma
->vm_mm
, address
, pmdp
);
429 /* need tlb flush only to serialize against gup-fast */
430 flush_tlb_range(vma
, address
, address
+ HPAGE_PMD_SIZE
);
436 * reserve_top_address - reserves a hole in the top of kernel address space
437 * @reserve - size of hole to reserve
439 * Can be used to relocate the fixmap area and poke a hole in the top
440 * of kernel address space to make room for a hypervisor.
442 void __init
reserve_top_address(unsigned long reserve
)
445 BUG_ON(fixmaps_set
> 0);
446 printk(KERN_INFO
"Reserving virtual address space above 0x%08x\n",
448 __FIXADDR_TOP
= -reserve
- PAGE_SIZE
;
454 void __native_set_fixmap(enum fixed_addresses idx
, pte_t pte
)
456 unsigned long address
= __fix_to_virt(idx
);
458 if (idx
>= __end_of_fixed_addresses
) {
462 set_pte_vaddr(address
, pte
);
466 void native_set_fixmap(enum fixed_addresses idx
, phys_addr_t phys
,
469 __native_set_fixmap(idx
, pfn_pte(phys
>> PAGE_SHIFT
, flags
));