| Chris Metcalf | 867e359 | 2010-05-28 23:09:12 -0400 | [diff] [blame] | 1 | /* | 
|  | 2 | * Copyright 2010 Tilera Corporation. All Rights Reserved. | 
|  | 3 | * | 
|  | 4 | *   This program is free software; you can redistribute it and/or | 
|  | 5 | *   modify it under the terms of the GNU General Public License | 
|  | 6 | *   as published by the Free Software Foundation, version 2. | 
|  | 7 | * | 
|  | 8 | *   This program is distributed in the hope that it will be useful, but | 
|  | 9 | *   WITHOUT ANY WARRANTY; without even the implied warranty of | 
|  | 10 | *   MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE, GOOD TITLE or | 
|  | 11 | *   NON INFRINGEMENT.  See the GNU General Public License for | 
|  | 12 | *   more details. | 
|  | 13 | * | 
|  | 14 | * TILE Huge TLB Page Support for Kernel. | 
|  | 15 | * Taken from i386 hugetlb implementation: | 
|  | 16 | * Copyright (C) 2002, Rohit Seth <rohit.seth@intel.com> | 
|  | 17 | */ | 
|  | 18 |  | 
|  | 19 | #include <linux/init.h> | 
|  | 20 | #include <linux/fs.h> | 
|  | 21 | #include <linux/mm.h> | 
|  | 22 | #include <linux/hugetlb.h> | 
|  | 23 | #include <linux/pagemap.h> | 
| Chris Metcalf | 867e359 | 2010-05-28 23:09:12 -0400 | [diff] [blame] | 24 | #include <linux/slab.h> | 
|  | 25 | #include <linux/err.h> | 
|  | 26 | #include <linux/sysctl.h> | 
|  | 27 | #include <linux/mman.h> | 
|  | 28 | #include <asm/tlb.h> | 
|  | 29 | #include <asm/tlbflush.h> | 
|  | 30 |  | 
|  | 31 | pte_t *huge_pte_alloc(struct mm_struct *mm, | 
|  | 32 | unsigned long addr, unsigned long sz) | 
|  | 33 | { | 
|  | 34 | pgd_t *pgd; | 
|  | 35 | pud_t *pud; | 
|  | 36 | pte_t *pte = NULL; | 
|  | 37 |  | 
|  | 38 | /* We do not yet support multiple huge page sizes. */ | 
|  | 39 | BUG_ON(sz != PMD_SIZE); | 
|  | 40 |  | 
|  | 41 | pgd = pgd_offset(mm, addr); | 
|  | 42 | pud = pud_alloc(mm, pgd, addr); | 
|  | 43 | if (pud) | 
|  | 44 | pte = (pte_t *) pmd_alloc(mm, pud, addr); | 
|  | 45 | BUG_ON(pte && !pte_none(*pte) && !pte_huge(*pte)); | 
|  | 46 |  | 
|  | 47 | return pte; | 
|  | 48 | } | 
|  | 49 |  | 
|  | 50 | pte_t *huge_pte_offset(struct mm_struct *mm, unsigned long addr) | 
|  | 51 | { | 
|  | 52 | pgd_t *pgd; | 
|  | 53 | pud_t *pud; | 
|  | 54 | pmd_t *pmd = NULL; | 
|  | 55 |  | 
|  | 56 | pgd = pgd_offset(mm, addr); | 
|  | 57 | if (pgd_present(*pgd)) { | 
|  | 58 | pud = pud_offset(pgd, addr); | 
|  | 59 | if (pud_present(*pud)) | 
|  | 60 | pmd = pmd_offset(pud, addr); | 
|  | 61 | } | 
|  | 62 | return (pte_t *) pmd; | 
|  | 63 | } | 
|  | 64 |  | 
|  | 65 | #ifdef HUGETLB_TEST | 
|  | 66 | struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address, | 
|  | 67 | int write) | 
|  | 68 | { | 
|  | 69 | unsigned long start = address; | 
|  | 70 | int length = 1; | 
|  | 71 | int nr; | 
|  | 72 | struct page *page; | 
|  | 73 | struct vm_area_struct *vma; | 
|  | 74 |  | 
|  | 75 | vma = find_vma(mm, addr); | 
|  | 76 | if (!vma || !is_vm_hugetlb_page(vma)) | 
|  | 77 | return ERR_PTR(-EINVAL); | 
|  | 78 |  | 
|  | 79 | pte = huge_pte_offset(mm, address); | 
|  | 80 |  | 
|  | 81 | /* hugetlb should be locked, and hence, prefaulted */ | 
|  | 82 | WARN_ON(!pte || pte_none(*pte)); | 
|  | 83 |  | 
|  | 84 | page = &pte_page(*pte)[vpfn % (HPAGE_SIZE/PAGE_SIZE)]; | 
|  | 85 |  | 
|  | 86 | WARN_ON(!PageHead(page)); | 
|  | 87 |  | 
|  | 88 | return page; | 
|  | 89 | } | 
|  | 90 |  | 
|  | 91 | int pmd_huge(pmd_t pmd) | 
|  | 92 | { | 
|  | 93 | return 0; | 
|  | 94 | } | 
|  | 95 |  | 
|  | 96 | int pud_huge(pud_t pud) | 
|  | 97 | { | 
|  | 98 | return 0; | 
|  | 99 | } | 
|  | 100 |  | 
|  | 101 | struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address, | 
|  | 102 | pmd_t *pmd, int write) | 
|  | 103 | { | 
|  | 104 | return NULL; | 
|  | 105 | } | 
|  | 106 |  | 
|  | 107 | #else | 
|  | 108 |  | 
|  | 109 | struct page *follow_huge_addr(struct mm_struct *mm, unsigned long address, | 
|  | 110 | int write) | 
|  | 111 | { | 
|  | 112 | return ERR_PTR(-EINVAL); | 
|  | 113 | } | 
|  | 114 |  | 
|  | 115 | int pmd_huge(pmd_t pmd) | 
|  | 116 | { | 
|  | 117 | return !!(pmd_val(pmd) & _PAGE_HUGE_PAGE); | 
|  | 118 | } | 
|  | 119 |  | 
|  | 120 | int pud_huge(pud_t pud) | 
|  | 121 | { | 
|  | 122 | return !!(pud_val(pud) & _PAGE_HUGE_PAGE); | 
|  | 123 | } | 
|  | 124 |  | 
|  | 125 | struct page *follow_huge_pmd(struct mm_struct *mm, unsigned long address, | 
|  | 126 | pmd_t *pmd, int write) | 
|  | 127 | { | 
|  | 128 | struct page *page; | 
|  | 129 |  | 
|  | 130 | page = pte_page(*(pte_t *)pmd); | 
|  | 131 | if (page) | 
|  | 132 | page += ((address & ~PMD_MASK) >> PAGE_SHIFT); | 
|  | 133 | return page; | 
|  | 134 | } | 
|  | 135 |  | 
|  | 136 | struct page *follow_huge_pud(struct mm_struct *mm, unsigned long address, | 
|  | 137 | pud_t *pud, int write) | 
|  | 138 | { | 
|  | 139 | struct page *page; | 
|  | 140 |  | 
|  | 141 | page = pte_page(*(pte_t *)pud); | 
|  | 142 | if (page) | 
|  | 143 | page += ((address & ~PUD_MASK) >> PAGE_SHIFT); | 
|  | 144 | return page; | 
|  | 145 | } | 
|  | 146 |  | 
|  | 147 | int huge_pmd_unshare(struct mm_struct *mm, unsigned long *addr, pte_t *ptep) | 
|  | 148 | { | 
|  | 149 | return 0; | 
|  | 150 | } | 
|  | 151 |  | 
|  | 152 | #endif | 
|  | 153 |  | 
|  | 154 | #ifdef HAVE_ARCH_HUGETLB_UNMAPPED_AREA | 
|  | 155 | static unsigned long hugetlb_get_unmapped_area_bottomup(struct file *file, | 
|  | 156 | unsigned long addr, unsigned long len, | 
|  | 157 | unsigned long pgoff, unsigned long flags) | 
|  | 158 | { | 
|  | 159 | struct hstate *h = hstate_file(file); | 
|  | 160 | struct mm_struct *mm = current->mm; | 
|  | 161 | struct vm_area_struct *vma; | 
|  | 162 | unsigned long start_addr; | 
|  | 163 |  | 
|  | 164 | if (len > mm->cached_hole_size) { | 
|  | 165 | start_addr = mm->free_area_cache; | 
|  | 166 | } else { | 
|  | 167 | start_addr = TASK_UNMAPPED_BASE; | 
|  | 168 | mm->cached_hole_size = 0; | 
|  | 169 | } | 
|  | 170 |  | 
|  | 171 | full_search: | 
|  | 172 | addr = ALIGN(start_addr, huge_page_size(h)); | 
|  | 173 |  | 
|  | 174 | for (vma = find_vma(mm, addr); ; vma = vma->vm_next) { | 
|  | 175 | /* At this point:  (!vma || addr < vma->vm_end). */ | 
|  | 176 | if (TASK_SIZE - len < addr) { | 
|  | 177 | /* | 
|  | 178 | * Start a new search - just in case we missed | 
|  | 179 | * some holes. | 
|  | 180 | */ | 
|  | 181 | if (start_addr != TASK_UNMAPPED_BASE) { | 
|  | 182 | start_addr = TASK_UNMAPPED_BASE; | 
|  | 183 | mm->cached_hole_size = 0; | 
|  | 184 | goto full_search; | 
|  | 185 | } | 
|  | 186 | return -ENOMEM; | 
|  | 187 | } | 
|  | 188 | if (!vma || addr + len <= vma->vm_start) { | 
|  | 189 | mm->free_area_cache = addr + len; | 
|  | 190 | return addr; | 
|  | 191 | } | 
|  | 192 | if (addr + mm->cached_hole_size < vma->vm_start) | 
|  | 193 | mm->cached_hole_size = vma->vm_start - addr; | 
|  | 194 | addr = ALIGN(vma->vm_end, huge_page_size(h)); | 
|  | 195 | } | 
|  | 196 | } | 
|  | 197 |  | 
|  | 198 | static unsigned long hugetlb_get_unmapped_area_topdown(struct file *file, | 
|  | 199 | unsigned long addr0, unsigned long len, | 
|  | 200 | unsigned long pgoff, unsigned long flags) | 
|  | 201 | { | 
|  | 202 | struct hstate *h = hstate_file(file); | 
|  | 203 | struct mm_struct *mm = current->mm; | 
|  | 204 | struct vm_area_struct *vma, *prev_vma; | 
|  | 205 | unsigned long base = mm->mmap_base, addr = addr0; | 
|  | 206 | unsigned long largest_hole = mm->cached_hole_size; | 
|  | 207 | int first_time = 1; | 
|  | 208 |  | 
|  | 209 | /* don't allow allocations above current base */ | 
|  | 210 | if (mm->free_area_cache > base) | 
|  | 211 | mm->free_area_cache = base; | 
|  | 212 |  | 
|  | 213 | if (len <= largest_hole) { | 
|  | 214 | largest_hole = 0; | 
|  | 215 | mm->free_area_cache  = base; | 
|  | 216 | } | 
|  | 217 | try_again: | 
|  | 218 | /* make sure it can fit in the remaining address space */ | 
|  | 219 | if (mm->free_area_cache < len) | 
|  | 220 | goto fail; | 
|  | 221 |  | 
|  | 222 | /* either no address requested or cant fit in requested address hole */ | 
|  | 223 | addr = (mm->free_area_cache - len) & huge_page_mask(h); | 
|  | 224 | do { | 
|  | 225 | /* | 
|  | 226 | * Lookup failure means no vma is above this address, | 
|  | 227 | * i.e. return with success: | 
|  | 228 | */ | 
|  | 229 | vma = find_vma_prev(mm, addr, &prev_vma); | 
|  | 230 | if (!vma) { | 
|  | 231 | return addr; | 
|  | 232 | break; | 
|  | 233 | } | 
|  | 234 |  | 
|  | 235 | /* | 
|  | 236 | * new region fits between prev_vma->vm_end and | 
|  | 237 | * vma->vm_start, use it: | 
|  | 238 | */ | 
|  | 239 | if (addr + len <= vma->vm_start && | 
|  | 240 | (!prev_vma || (addr >= prev_vma->vm_end))) { | 
|  | 241 | /* remember the address as a hint for next time */ | 
|  | 242 | mm->cached_hole_size = largest_hole; | 
|  | 243 | mm->free_area_cache = addr; | 
|  | 244 | return addr; | 
|  | 245 | } else { | 
|  | 246 | /* pull free_area_cache down to the first hole */ | 
|  | 247 | if (mm->free_area_cache == vma->vm_end) { | 
|  | 248 | mm->free_area_cache = vma->vm_start; | 
|  | 249 | mm->cached_hole_size = largest_hole; | 
|  | 250 | } | 
|  | 251 | } | 
|  | 252 |  | 
|  | 253 | /* remember the largest hole we saw so far */ | 
|  | 254 | if (addr + largest_hole < vma->vm_start) | 
|  | 255 | largest_hole = vma->vm_start - addr; | 
|  | 256 |  | 
|  | 257 | /* try just below the current vma->vm_start */ | 
|  | 258 | addr = (vma->vm_start - len) & huge_page_mask(h); | 
|  | 259 |  | 
|  | 260 | } while (len <= vma->vm_start); | 
|  | 261 |  | 
|  | 262 | fail: | 
|  | 263 | /* | 
|  | 264 | * if hint left us with no space for the requested | 
|  | 265 | * mapping then try again: | 
|  | 266 | */ | 
|  | 267 | if (first_time) { | 
|  | 268 | mm->free_area_cache = base; | 
|  | 269 | largest_hole = 0; | 
|  | 270 | first_time = 0; | 
|  | 271 | goto try_again; | 
|  | 272 | } | 
|  | 273 | /* | 
|  | 274 | * A failed mmap() very likely causes application failure, | 
|  | 275 | * so fall back to the bottom-up function here. This scenario | 
|  | 276 | * can happen with large stack limits and large mmap() | 
|  | 277 | * allocations. | 
|  | 278 | */ | 
|  | 279 | mm->free_area_cache = TASK_UNMAPPED_BASE; | 
|  | 280 | mm->cached_hole_size = ~0UL; | 
|  | 281 | addr = hugetlb_get_unmapped_area_bottomup(file, addr0, | 
|  | 282 | len, pgoff, flags); | 
|  | 283 |  | 
|  | 284 | /* | 
|  | 285 | * Restore the topdown base: | 
|  | 286 | */ | 
|  | 287 | mm->free_area_cache = base; | 
|  | 288 | mm->cached_hole_size = ~0UL; | 
|  | 289 |  | 
|  | 290 | return addr; | 
|  | 291 | } | 
|  | 292 |  | 
|  | 293 | unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr, | 
|  | 294 | unsigned long len, unsigned long pgoff, unsigned long flags) | 
|  | 295 | { | 
|  | 296 | struct hstate *h = hstate_file(file); | 
|  | 297 | struct mm_struct *mm = current->mm; | 
|  | 298 | struct vm_area_struct *vma; | 
|  | 299 |  | 
|  | 300 | if (len & ~huge_page_mask(h)) | 
|  | 301 | return -EINVAL; | 
|  | 302 | if (len > TASK_SIZE) | 
|  | 303 | return -ENOMEM; | 
|  | 304 |  | 
|  | 305 | if (flags & MAP_FIXED) { | 
|  | 306 | if (prepare_hugepage_range(file, addr, len)) | 
|  | 307 | return -EINVAL; | 
|  | 308 | return addr; | 
|  | 309 | } | 
|  | 310 |  | 
|  | 311 | if (addr) { | 
|  | 312 | addr = ALIGN(addr, huge_page_size(h)); | 
|  | 313 | vma = find_vma(mm, addr); | 
|  | 314 | if (TASK_SIZE - len >= addr && | 
|  | 315 | (!vma || addr + len <= vma->vm_start)) | 
|  | 316 | return addr; | 
|  | 317 | } | 
|  | 318 | if (current->mm->get_unmapped_area == arch_get_unmapped_area) | 
|  | 319 | return hugetlb_get_unmapped_area_bottomup(file, addr, len, | 
|  | 320 | pgoff, flags); | 
|  | 321 | else | 
|  | 322 | return hugetlb_get_unmapped_area_topdown(file, addr, len, | 
|  | 323 | pgoff, flags); | 
|  | 324 | } | 
|  | 325 |  | 
|  | 326 | static __init int setup_hugepagesz(char *opt) | 
|  | 327 | { | 
|  | 328 | unsigned long ps = memparse(opt, &opt); | 
|  | 329 | if (ps == PMD_SIZE) { | 
|  | 330 | hugetlb_add_hstate(PMD_SHIFT - PAGE_SHIFT); | 
|  | 331 | } else if (ps == PUD_SIZE) { | 
|  | 332 | hugetlb_add_hstate(PUD_SHIFT - PAGE_SHIFT); | 
|  | 333 | } else { | 
| Chris Metcalf | 0707ad3 | 2010-06-25 17:04:17 -0400 | [diff] [blame] | 334 | pr_err("hugepagesz: Unsupported page size %lu M\n", | 
| Chris Metcalf | 867e359 | 2010-05-28 23:09:12 -0400 | [diff] [blame] | 335 | ps >> 20); | 
|  | 336 | return 0; | 
|  | 337 | } | 
|  | 338 | return 1; | 
|  | 339 | } | 
|  | 340 | __setup("hugepagesz=", setup_hugepagesz); | 
|  | 341 |  | 
|  | 342 | #endif /*HAVE_ARCH_HUGETLB_UNMAPPED_AREA*/ |