diff options
author | David Gibson <david@gibson.dropbear.id.au> | 2005-06-21 20:14:44 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-06-21 21:46:15 -0400 |
commit | 63551ae0feaaa23807ebea60de1901564bbef32e (patch) | |
tree | f6f97f60f83c3e9813bdfcc6039c499997b1ea10 /arch/sparc64 | |
parent | 1e7e5a9048b30c57ba1ddaa6cdf59b21b65cde99 (diff) |
[PATCH] Hugepage consolidation
A lot of the code in arch/*/mm/hugetlbpage.c is quite similar. This patch
attempts to consolidate a lot of the code across the arch's, putting the
combined version in mm/hugetlb.c. There are a couple of uglyish hacks in
order to covert all the hugepage archs, but the result is a very large
reduction in the total amount of code. It also means things like hugepage
lazy allocation could be implemented in one place, instead of six.
Tested, at least a little, on ppc64, i386 and x86_64.
Notes:
- this patch changes the meaning of set_huge_pte() to be more
analagous to set_pte()
- does SH4 need s special huge_ptep_get_and_clear()??
Acked-by: William Lee Irwin <wli@holomorphy.com>
Signed-off-by: Andrew Morton <akpm@osdl.org>
Signed-off-by: Linus Torvalds <torvalds@osdl.org>
Diffstat (limited to 'arch/sparc64')
-rw-r--r-- | arch/sparc64/mm/hugetlbpage.c | 195 |
1 files changed, 24 insertions, 171 deletions
diff --git a/arch/sparc64/mm/hugetlbpage.c b/arch/sparc64/mm/hugetlbpage.c index 5a1f831b2de..625cbb336a2 100644 --- a/arch/sparc64/mm/hugetlbpage.c +++ b/arch/sparc64/mm/hugetlbpage.c | |||
@@ -22,7 +22,7 @@ | |||
22 | #include <asm/cacheflush.h> | 22 | #include <asm/cacheflush.h> |
23 | #include <asm/mmu_context.h> | 23 | #include <asm/mmu_context.h> |
24 | 24 | ||
25 | static pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr) | 25 | pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr) |
26 | { | 26 | { |
27 | pgd_t *pgd; | 27 | pgd_t *pgd; |
28 | pud_t *pud; | 28 | pud_t *pud; |
@@ -41,7 +41,7 @@ static pte_t *huge_pte_alloc(struct mm_struct *mm, unsigned long addr) | |||
41 | return pte; | 41 | return pte; |
42 | } | 42 | } |
43 | 43 | ||
44 | static pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr) | 44 | pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr) |
45 | { | 45 | { |
46 | pgd_t *pgd; | 46 | pgd_t *pgd; |
47 | pud_t *pud; | 47 | pud_t *pud; |
@@ -62,30 +62,34 @@ static pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr) | |||
62 | 62 | ||
63 | #define mk_pte_huge(entry) do { pte_val(entry) |= _PAGE_SZHUGE; } while (0) | 63 | #define mk_pte_huge(entry) do { pte_val(entry) |= _PAGE_SZHUGE; } while (0) |
64 | 64 | ||
65 | static void set_huge_pte(struct mm_struct *mm, struct vm_area_struct *vma, | 65 | void set_huge_pte_at(struct mm_struct *mm, unsigned long addr, |
66 | unsigned long addr, | 66 | pte_t *ptep, pte_t entry) |
67 | struct page *page, pte_t * page_table, int write_access) | ||
68 | { | 67 | { |
69 | unsigned long i; | 68 | int i; |
70 | pte_t entry; | 69 | |
70 | for (i = 0; i < (1 << HUGETLB_PAGE_ORDER); i++) { | ||
71 | set_pte_at(mm, addr, ptep, entry); | ||
72 | ptep++; | ||
73 | addr += PAGE_SIZE; | ||
74 | pte_val(entry) += PAGE_SIZE; | ||
75 | } | ||
76 | } | ||
71 | 77 | ||
72 | add_mm_counter(mm, rss, HPAGE_SIZE / PAGE_SIZE); | 78 | pte_t huge_ptep_get_and_clear(struct mm_struct *mm, unsigned long addr, |
79 | pte_t *ptep) | ||
80 | { | ||
81 | pte_t entry; | ||
82 | int i; | ||
73 | 83 | ||
74 | if (write_access) | 84 | entry = *ptep; |
75 | entry = pte_mkwrite(pte_mkdirty(mk_pte(page, | ||
76 | vma->vm_page_prot))); | ||
77 | else | ||
78 | entry = pte_wrprotect(mk_pte(page, vma->vm_page_prot)); | ||
79 | entry = pte_mkyoung(entry); | ||
80 | mk_pte_huge(entry); | ||
81 | 85 | ||
82 | for (i = 0; i < (1 << HUGETLB_PAGE_ORDER); i++) { | 86 | for (i = 0; i < (1 << HUGETLB_PAGE_ORDER); i++) { |
83 | set_pte_at(mm, addr, page_table, entry); | 87 | pte_clear(mm, addr, ptep); |
84 | page_table++; | ||
85 | addr += PAGE_SIZE; | 88 | addr += PAGE_SIZE; |
86 | 89 | ptep++; | |
87 | pte_val(entry) += PAGE_SIZE; | ||
88 | } | 90 | } |
91 | |||
92 | return entry; | ||
89 | } | 93 | } |
90 | 94 | ||
91 | /* | 95 | /* |
@@ -100,79 +104,6 @@ int is_aligned_hugepage_range(unsigned long addr, unsigned long len) | |||
100 | return 0; | 104 | return 0; |
101 | } | 105 | } |
102 | 106 | ||
103 | int copy_hugetlb_page_range(struct mm_struct *dst, struct mm_struct *src, | ||
104 | struct vm_area_struct *vma) | ||
105 | { | ||
106 | pte_t *src_pte, *dst_pte, entry; | ||
107 | struct page *ptepage; | ||
108 | unsigned long addr = vma->vm_start; | ||
109 | unsigned long end = vma->vm_end; | ||
110 | int i; | ||
111 | |||
112 | while (addr < end) { | ||
113 | dst_pte = huge_pte_alloc(dst, addr); | ||
114 | if (!dst_pte) | ||
115 | goto nomem; | ||
116 | src_pte = huge_pte_offset(src, addr); | ||
117 | BUG_ON(!src_pte || pte_none(*src_pte)); | ||
118 | entry = *src_pte; | ||
119 | ptepage = pte_page(entry); | ||
120 | get_page(ptepage); | ||
121 | for (i = 0; i < (1 << HUGETLB_PAGE_ORDER); i++) { | ||
122 | set_pte_at(dst, addr, dst_pte, entry); | ||
123 | pte_val(entry) += PAGE_SIZE; | ||
124 | dst_pte++; | ||
125 | addr += PAGE_SIZE; | ||
126 | } | ||
127 | add_mm_counter(dst, rss, HPAGE_SIZE / PAGE_SIZE); | ||
128 | } | ||
129 | return 0; | ||
130 | |||
131 | nomem: | ||
132 | return -ENOMEM; | ||
133 | } | ||
134 | |||
135 | int follow_hugetlb_page(struct mm_struct *mm, struct vm_area_struct *vma, | ||
136 | struct page **pages, struct vm_area_struct **vmas, | ||
137 | unsigned long *position, int *length, int i) | ||
138 | { | ||
139 | unsigned long vaddr = *position; | ||
140 | int remainder = *length; | ||
141 | |||
142 | WARN_ON(!is_vm_hugetlb_page(vma)); | ||
143 | |||
144 | while (vaddr < vma->vm_end && remainder) { | ||
145 | if (pages) { | ||
146 | pte_t *pte; | ||
147 | struct page *page; | ||
148 | |||
149 | pte = huge_pte_offset(mm, vaddr); | ||
150 | |||
151 | /* hugetlb should be locked, and hence, prefaulted */ | ||
152 | BUG_ON(!pte || pte_none(*pte)); | ||
153 | |||
154 | page = pte_page(*pte); | ||
155 | |||
156 | WARN_ON(!PageCompound(page)); | ||
157 | |||
158 | get_page(page); | ||
159 | pages[i] = page; | ||
160 | } | ||
161 | |||
162 | if (vmas) | ||
163 | vmas[i] = vma; | ||
164 | |||
165 | vaddr += PAGE_SIZE; | ||
166 | --remainder; | ||
167 | ++i; | ||
168 | } | ||
169 | |||
170 | *length = remainder; | ||
171 | *position = vaddr; | ||
172 | |||
173 | return i; | ||
174 | } | ||
175 | |||
176 | struct page *follow_huge_addr(struct mm_struct *mm, | 107 | struct page *follow_huge_addr(struct mm_struct *mm, |
177 | unsigned long address, int write) | 108 | unsigned long address, int write) |
178 | { | 109 | { |
@@ -190,34 +121,6 @@ struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address, | |||
190 | return NULL; | 121 | return NULL; |
191 | } | 122 | } |
192 | 123 | ||
193 | void unmap_hugepage_range(struct vm_area_struct *vma, | ||
194 | unsigned long start, unsigned long end) | ||
195 | { | ||
196 | struct mm_struct *mm = vma->vm_mm; | ||
197 | unsigned long address; | ||
198 | pte_t *pte; | ||
199 | struct page *page; | ||
200 | int i; | ||
201 | |||
202 | BUG_ON(start & (HPAGE_SIZE - 1)); | ||
203 | BUG_ON(end & (HPAGE_SIZE - 1)); | ||
204 | |||
205 | for (address = start; address < end; address += HPAGE_SIZE) { | ||
206 | pte = huge_pte_offset(mm, address); | ||
207 | BUG_ON(!pte); | ||
208 | if (pte_none(*pte)) | ||
209 | continue; | ||
210 | page = pte_page(*pte); | ||
211 | put_page(page); | ||
212 | for (i = 0; i < (1 << HUGETLB_PAGE_ORDER); i++) { | ||
213 | pte_clear(mm, address+(i*PAGE_SIZE), pte); | ||
214 | pte++; | ||
215 | } | ||
216 | } | ||
217 | add_mm_counter(mm, rss, -((end - start) >> PAGE_SHIFT)); | ||
218 | flush_tlb_range(vma, start, end); | ||
219 | } | ||
220 | |||
221 | static void context_reload(void *__data) | 124 | static void context_reload(void *__data) |
222 | { | 125 | { |
223 | struct mm_struct *mm = __data; | 126 | struct mm_struct *mm = __data; |
@@ -226,12 +129,8 @@ static void context_reload(void *__data) | |||
226 | load_secondary_context(mm); | 129 | load_secondary_context(mm); |
227 | } | 130 | } |
228 | 131 | ||
229 | int hugetlb_prefault(struct address_space *mapping, struct vm_area_struct *vma) | 132 | void hugetlb_prefault_arch_hook(struct mm_struct *mm) |
230 | { | 133 | { |
231 | struct mm_struct *mm = current->mm; | ||
232 | unsigned long addr; | ||
233 | int ret = 0; | ||
234 | |||
235 | /* On UltraSPARC-III+ and later, configure the second half of | 134 | /* On UltraSPARC-III+ and later, configure the second half of |
236 | * the Data-TLB for huge pages. | 135 | * the Data-TLB for huge pages. |
237 | */ | 136 | */ |
@@ -261,50 +160,4 @@ int hugetlb_prefault(struct address_space *mapping, struct vm_area_struct *vma) | |||
261 | } | 160 | } |
262 | spin_unlock(&ctx_alloc_lock); | 161 | spin_unlock(&ctx_alloc_lock); |
263 | } | 162 | } |
264 | |||
265 | BUG_ON(vma->vm_start & ~HPAGE_MASK); | ||
266 | BUG_ON(vma->vm_end & ~HPAGE_MASK); | ||
267 | |||
268 | spin_lock(&mm->page_table_lock); | ||
269 | for (addr = vma->vm_start; addr < vma->vm_end; addr += HPAGE_SIZE) { | ||
270 | unsigned long idx; | ||
271 | pte_t *pte = huge_pte_alloc(mm, addr); | ||
272 | struct page *page; | ||
273 | |||
274 | if (!pte) { | ||
275 | ret = -ENOMEM; | ||
276 | goto out; | ||
277 | } | ||
278 | if (!pte_none(*pte)) | ||
279 | continue; | ||
280 | |||
281 | idx = ((addr - vma->vm_start) >> HPAGE_SHIFT) | ||
282 | + (vma->vm_pgoff >> (HPAGE_SHIFT - PAGE_SHIFT)); | ||
283 | page = find_get_page(mapping, idx); | ||
284 | if (!page) { | ||
285 | /* charge the fs quota first */ | ||
286 | if (hugetlb_get_quota(mapping)) { | ||
287 | ret = -ENOMEM; | ||
288 | goto out; | ||
289 | } | ||
290 | page = alloc_huge_page(); | ||
291 | if (!page) { | ||
292 | hugetlb_put_quota(mapping); | ||
293 | ret = -ENOMEM; | ||
294 | goto out; | ||
295 | } | ||
296 | ret = add_to_page_cache(page, mapping, idx, GFP_ATOMIC); | ||
297 | if (! ret) { | ||
298 | unlock_page(page); | ||
299 | } else { | ||
300 | hugetlb_put_quota(mapping); | ||
301 | free_huge_page(page); | ||
302 | goto out; | ||
303 | } | ||
304 | } | ||
305 | set_huge_pte(mm, vma, addr, page, pte, vma->vm_flags & VM_WRITE); | ||
306 | } | ||
307 | out: | ||
308 | spin_unlock(&mm->page_table_lock); | ||
309 | return ret; | ||
310 | } | 163 | } |