2 * IA-64 Huge TLB Page Support for Kernel.
4 * Copyright (C) 2002-2004 Rohit Seth <rohit.seth@intel.com>
5 * Copyright (C) 2003-2004 Ken Chen <kenneth.w.chen@intel.com>
7 * Sep, 2003: add numa support
8 * Feb, 2004: dynamic hugetlb page size via boot parameter
11 #include <linux/init.h>
14 #include <linux/hugetlb.h>
15 #include <linux/pagemap.h>
16 #include <linux/module.h>
17 #include <linux/sysctl.h>
18 #include <linux/log2.h>
20 #include <asm/pgalloc.h>
22 #include <asm/tlbflush.h>
24 unsigned int hpage_shift
= HPAGE_SHIFT_DEFAULT
;
25 EXPORT_SYMBOL(hpage_shift
);
28 huge_pte_alloc(struct mm_struct
*mm
, unsigned long addr
, unsigned long sz
)
30 unsigned long taddr
= htlbpage_to_page(addr
);
36 pgd
= pgd_offset(mm
, taddr
);
37 pud
= pud_alloc(mm
, pgd
, taddr
);
39 pmd
= pmd_alloc(mm
, pud
, taddr
);
41 pte
= pte_alloc_map(mm
, NULL
, pmd
, taddr
);
47 huge_pte_offset (struct mm_struct
*mm
, unsigned long addr
)
49 unsigned long taddr
= htlbpage_to_page(addr
);
55 pgd
= pgd_offset(mm
, taddr
);
56 if (pgd_present(*pgd
)) {
57 pud
= pud_offset(pgd
, taddr
);
58 if (pud_present(*pud
)) {
59 pmd
= pmd_offset(pud
, taddr
);
60 if (pmd_present(*pmd
))
61 pte
= pte_offset_map(pmd
, taddr
);
68 int huge_pmd_unshare(struct mm_struct
*mm
, unsigned long *addr
, pte_t
*ptep
)
73 #define mk_pte_huge(entry) { pte_val(entry) |= _PAGE_P; }
76 * Don't actually need to do any preparation, but need to make sure
77 * the address is in the right region.
79 int prepare_hugepage_range(struct file
*file
,
80 unsigned long addr
, unsigned long len
)
82 if (len
& ~HPAGE_MASK
)
84 if (addr
& ~HPAGE_MASK
)
86 if (REGION_NUMBER(addr
) != RGN_HPAGE
)
92 struct page
*follow_huge_addr(struct mm_struct
*mm
, unsigned long addr
, int write
)
97 if (REGION_NUMBER(addr
) != RGN_HPAGE
)
98 return ERR_PTR(-EINVAL
);
100 ptep
= huge_pte_offset(mm
, addr
);
101 if (!ptep
|| pte_none(*ptep
))
103 page
= pte_page(*ptep
);
104 page
+= ((addr
& ~HPAGE_MASK
) >> PAGE_SHIFT
);
107 int pmd_huge(pmd_t pmd
)
112 int pud_huge(pud_t pud
)
118 follow_huge_pmd(struct mm_struct
*mm
, unsigned long address
, pmd_t
*pmd
, int write
)
123 void hugetlb_free_pgd_range(struct mmu_gather
*tlb
,
124 unsigned long addr
, unsigned long end
,
125 unsigned long floor
, unsigned long ceiling
)
128 * This is called to free hugetlb page tables.
130 * The offset of these addresses from the base of the hugetlb
131 * region must be scaled down by HPAGE_SIZE/PAGE_SIZE so that
132 * the standard free_pgd_range will free the right page tables.
134 * If floor and ceiling are also in the hugetlb region, they
135 * must likewise be scaled down; but if outside, left unchanged.
138 addr
= htlbpage_to_page(addr
);
139 end
= htlbpage_to_page(end
);
140 if (REGION_NUMBER(floor
) == RGN_HPAGE
)
141 floor
= htlbpage_to_page(floor
);
142 if (REGION_NUMBER(ceiling
) == RGN_HPAGE
)
143 ceiling
= htlbpage_to_page(ceiling
);
145 free_pgd_range(tlb
, addr
, end
, floor
, ceiling
);
148 unsigned long hugetlb_get_unmapped_area(struct file
*file
, unsigned long addr
, unsigned long len
,
149 unsigned long pgoff
, unsigned long flags
)
151 struct vm_unmapped_area_info info
;
153 if (len
> RGN_MAP_LIMIT
)
155 if (len
& ~HPAGE_MASK
)
158 /* Handle MAP_FIXED */
159 if (flags
& MAP_FIXED
) {
160 if (prepare_hugepage_range(file
, addr
, len
))
165 /* This code assumes that RGN_HPAGE != 0. */
166 if ((REGION_NUMBER(addr
) != RGN_HPAGE
) || (addr
& (HPAGE_SIZE
- 1)))
167 addr
= HPAGE_REGION_BASE
;
171 info
.low_limit
= addr
;
172 info
.high_limit
= HPAGE_REGION_BASE
+ RGN_MAP_LIMIT
;
173 info
.align_mask
= PAGE_MASK
& (HPAGE_SIZE
- 1);
174 info
.align_offset
= 0;
175 return vm_unmapped_area(&info
);
178 static int __init
hugetlb_setup_sz(char *str
)
181 unsigned long long size
;
183 if (ia64_pal_vm_page_size(&tr_pages
, NULL
) != 0)
185 * shouldn't happen, but just in case.
187 tr_pages
= 0x15557000UL
;
189 size
= memparse(str
, &str
);
190 if (*str
|| !is_power_of_2(size
) || !(tr_pages
& size
) ||
192 size
>= (1UL << PAGE_SHIFT
<< MAX_ORDER
)) {
193 printk(KERN_WARNING
"Invalid huge page size specified\n");
197 hpage_shift
= __ffs(size
);
199 * boot cpu already executed ia64_mmu_init, and has HPAGE_SHIFT_DEFAULT
200 * override here with new page shift.
202 ia64_set_rr(HPAGE_REGION_BASE
, hpage_shift
<< 2);
205 early_param("hugepagesz", hugetlb_setup_sz
);