Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 1 | /* |
| 2 | * address space "slices" (meta-segments) support |
| 3 | * |
| 4 | * Copyright (C) 2007 Benjamin Herrenschmidt, IBM Corporation. |
| 5 | * |
| 6 | * Based on hugetlb implementation |
| 7 | * |
| 8 | * Copyright (C) 2003 David Gibson, IBM Corporation. |
| 9 | * |
| 10 | * This program is free software; you can redistribute it and/or modify |
| 11 | * it under the terms of the GNU General Public License as published by |
| 12 | * the Free Software Foundation; either version 2 of the License, or |
| 13 | * (at your option) any later version. |
| 14 | * |
| 15 | * This program is distributed in the hope that it will be useful, |
| 16 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 17 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 18 | * GNU General Public License for more details. |
| 19 | * |
| 20 | * You should have received a copy of the GNU General Public License |
| 21 | * along with this program; if not, write to the Free Software |
| 22 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 23 | */ |
| 24 | |
| 25 | #undef DEBUG |
| 26 | |
| 27 | #include <linux/kernel.h> |
| 28 | #include <linux/mm.h> |
| 29 | #include <linux/pagemap.h> |
| 30 | #include <linux/err.h> |
| 31 | #include <linux/spinlock.h> |
| 32 | #include <linux/export.h> |
| 33 | #include <linux/hugetlb.h> |
| 34 | #include <asm/mman.h> |
| 35 | #include <asm/mmu.h> |
| 36 | #include <asm/copro.h> |
| 37 | #include <asm/hugetlb.h> |
| 38 | #include <asm/mmu_context.h> |
| 39 | |
| 40 | static DEFINE_SPINLOCK(slice_convert_lock); |
| 41 | |
| 42 | #ifdef DEBUG |
| 43 | int _slice_debug = 1; |
| 44 | |
| 45 | static void slice_print_mask(const char *label, const struct slice_mask *mask) |
| 46 | { |
| 47 | if (!_slice_debug) |
| 48 | return; |
| 49 | pr_devel("%s low_slice: %*pbl\n", label, |
| 50 | (int)SLICE_NUM_LOW, &mask->low_slices); |
| 51 | pr_devel("%s high_slice: %*pbl\n", label, |
| 52 | (int)SLICE_NUM_HIGH, mask->high_slices); |
| 53 | } |
| 54 | |
| 55 | #define slice_dbg(fmt...) do { if (_slice_debug) pr_devel(fmt); } while (0) |
| 56 | |
| 57 | #else |
| 58 | |
| 59 | static void slice_print_mask(const char *label, const struct slice_mask *mask) {} |
| 60 | #define slice_dbg(fmt...) |
| 61 | |
| 62 | #endif |
| 63 | |
| 64 | static inline bool slice_addr_is_low(unsigned long addr) |
| 65 | { |
| 66 | u64 tmp = (u64)addr; |
| 67 | |
| 68 | return tmp < SLICE_LOW_TOP; |
| 69 | } |
| 70 | |
| 71 | static void slice_range_to_mask(unsigned long start, unsigned long len, |
| 72 | struct slice_mask *ret) |
| 73 | { |
| 74 | unsigned long end = start + len - 1; |
| 75 | |
| 76 | ret->low_slices = 0; |
| 77 | if (SLICE_NUM_HIGH) |
| 78 | bitmap_zero(ret->high_slices, SLICE_NUM_HIGH); |
| 79 | |
| 80 | if (slice_addr_is_low(start)) { |
| 81 | unsigned long mend = min(end, |
| 82 | (unsigned long)(SLICE_LOW_TOP - 1)); |
| 83 | |
| 84 | ret->low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1)) |
| 85 | - (1u << GET_LOW_SLICE_INDEX(start)); |
| 86 | } |
| 87 | |
| 88 | if (SLICE_NUM_HIGH && !slice_addr_is_low(end)) { |
| 89 | unsigned long start_index = GET_HIGH_SLICE_INDEX(start); |
| 90 | unsigned long align_end = ALIGN(end, (1UL << SLICE_HIGH_SHIFT)); |
| 91 | unsigned long count = GET_HIGH_SLICE_INDEX(align_end) - start_index; |
| 92 | |
| 93 | bitmap_set(ret->high_slices, start_index, count); |
| 94 | } |
| 95 | } |
| 96 | |
| 97 | static int slice_area_is_free(struct mm_struct *mm, unsigned long addr, |
| 98 | unsigned long len) |
| 99 | { |
| 100 | struct vm_area_struct *vma; |
| 101 | |
| 102 | if ((mm->context.slb_addr_limit - len) < addr) |
| 103 | return 0; |
| 104 | vma = find_vma(mm, addr); |
| 105 | return (!vma || (addr + len) <= vm_start_gap(vma)); |
| 106 | } |
| 107 | |
| 108 | static int slice_low_has_vma(struct mm_struct *mm, unsigned long slice) |
| 109 | { |
| 110 | return !slice_area_is_free(mm, slice << SLICE_LOW_SHIFT, |
| 111 | 1ul << SLICE_LOW_SHIFT); |
| 112 | } |
| 113 | |
| 114 | static int slice_high_has_vma(struct mm_struct *mm, unsigned long slice) |
| 115 | { |
| 116 | unsigned long start = slice << SLICE_HIGH_SHIFT; |
| 117 | unsigned long end = start + (1ul << SLICE_HIGH_SHIFT); |
| 118 | |
| 119 | #ifdef CONFIG_PPC64 |
| 120 | /* Hack, so that each addresses is controlled by exactly one |
| 121 | * of the high or low area bitmaps, the first high area starts |
| 122 | * at 4GB, not 0 */ |
| 123 | if (start == 0) |
| 124 | start = SLICE_LOW_TOP; |
| 125 | #endif |
| 126 | |
| 127 | return !slice_area_is_free(mm, start, end - start); |
| 128 | } |
| 129 | |
| 130 | static void slice_mask_for_free(struct mm_struct *mm, struct slice_mask *ret, |
| 131 | unsigned long high_limit) |
| 132 | { |
| 133 | unsigned long i; |
| 134 | |
| 135 | ret->low_slices = 0; |
| 136 | if (SLICE_NUM_HIGH) |
| 137 | bitmap_zero(ret->high_slices, SLICE_NUM_HIGH); |
| 138 | |
| 139 | for (i = 0; i < SLICE_NUM_LOW; i++) |
| 140 | if (!slice_low_has_vma(mm, i)) |
| 141 | ret->low_slices |= 1u << i; |
| 142 | |
| 143 | if (slice_addr_is_low(high_limit - 1)) |
| 144 | return; |
| 145 | |
| 146 | for (i = 0; i < GET_HIGH_SLICE_INDEX(high_limit); i++) |
| 147 | if (!slice_high_has_vma(mm, i)) |
| 148 | __set_bit(i, ret->high_slices); |
| 149 | } |
| 150 | |
| 151 | #ifdef CONFIG_PPC_BOOK3S_64 |
| 152 | static struct slice_mask *slice_mask_for_size(struct mm_struct *mm, int psize) |
| 153 | { |
| 154 | #ifdef CONFIG_PPC_64K_PAGES |
| 155 | if (psize == MMU_PAGE_64K) |
| 156 | return &mm->context.mask_64k; |
| 157 | #endif |
| 158 | if (psize == MMU_PAGE_4K) |
| 159 | return &mm->context.mask_4k; |
| 160 | #ifdef CONFIG_HUGETLB_PAGE |
| 161 | if (psize == MMU_PAGE_16M) |
| 162 | return &mm->context.mask_16m; |
| 163 | if (psize == MMU_PAGE_16G) |
| 164 | return &mm->context.mask_16g; |
| 165 | #endif |
| 166 | BUG(); |
| 167 | } |
| 168 | #elif defined(CONFIG_PPC_8xx) |
| 169 | static struct slice_mask *slice_mask_for_size(struct mm_struct *mm, int psize) |
| 170 | { |
| 171 | if (psize == mmu_virtual_psize) |
| 172 | return &mm->context.mask_base_psize; |
| 173 | #ifdef CONFIG_HUGETLB_PAGE |
| 174 | if (psize == MMU_PAGE_512K) |
| 175 | return &mm->context.mask_512k; |
| 176 | if (psize == MMU_PAGE_8M) |
| 177 | return &mm->context.mask_8m; |
| 178 | #endif |
| 179 | BUG(); |
| 180 | } |
| 181 | #else |
| 182 | #error "Must define the slice masks for page sizes supported by the platform" |
| 183 | #endif |
| 184 | |
| 185 | static bool slice_check_range_fits(struct mm_struct *mm, |
| 186 | const struct slice_mask *available, |
| 187 | unsigned long start, unsigned long len) |
| 188 | { |
| 189 | unsigned long end = start + len - 1; |
| 190 | u64 low_slices = 0; |
| 191 | |
| 192 | if (slice_addr_is_low(start)) { |
| 193 | unsigned long mend = min(end, |
| 194 | (unsigned long)(SLICE_LOW_TOP - 1)); |
| 195 | |
| 196 | low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1)) |
| 197 | - (1u << GET_LOW_SLICE_INDEX(start)); |
| 198 | } |
| 199 | if ((low_slices & available->low_slices) != low_slices) |
| 200 | return false; |
| 201 | |
| 202 | if (SLICE_NUM_HIGH && !slice_addr_is_low(end)) { |
| 203 | unsigned long start_index = GET_HIGH_SLICE_INDEX(start); |
| 204 | unsigned long align_end = ALIGN(end, (1UL << SLICE_HIGH_SHIFT)); |
| 205 | unsigned long count = GET_HIGH_SLICE_INDEX(align_end) - start_index; |
| 206 | unsigned long i; |
| 207 | |
| 208 | for (i = start_index; i < start_index + count; i++) { |
| 209 | if (!test_bit(i, available->high_slices)) |
| 210 | return false; |
| 211 | } |
| 212 | } |
| 213 | |
| 214 | return true; |
| 215 | } |
| 216 | |
| 217 | static void slice_flush_segments(void *parm) |
| 218 | { |
| 219 | #ifdef CONFIG_PPC64 |
| 220 | struct mm_struct *mm = parm; |
| 221 | unsigned long flags; |
| 222 | |
| 223 | if (mm != current->active_mm) |
| 224 | return; |
| 225 | |
| 226 | copy_mm_to_paca(current->active_mm); |
| 227 | |
| 228 | local_irq_save(flags); |
| 229 | slb_flush_and_rebolt(); |
| 230 | local_irq_restore(flags); |
| 231 | #endif |
| 232 | } |
| 233 | |
| 234 | static void slice_convert(struct mm_struct *mm, |
| 235 | const struct slice_mask *mask, int psize) |
| 236 | { |
| 237 | int index, mask_index; |
| 238 | /* Write the new slice psize bits */ |
| 239 | unsigned char *hpsizes, *lpsizes; |
| 240 | struct slice_mask *psize_mask, *old_mask; |
| 241 | unsigned long i, flags; |
| 242 | int old_psize; |
| 243 | |
| 244 | slice_dbg("slice_convert(mm=%p, psize=%d)\n", mm, psize); |
| 245 | slice_print_mask(" mask", mask); |
| 246 | |
| 247 | psize_mask = slice_mask_for_size(mm, psize); |
| 248 | |
| 249 | /* We need to use a spinlock here to protect against |
| 250 | * concurrent 64k -> 4k demotion ... |
| 251 | */ |
| 252 | spin_lock_irqsave(&slice_convert_lock, flags); |
| 253 | |
| 254 | lpsizes = mm->context.low_slices_psize; |
| 255 | for (i = 0; i < SLICE_NUM_LOW; i++) { |
| 256 | if (!(mask->low_slices & (1u << i))) |
| 257 | continue; |
| 258 | |
| 259 | mask_index = i & 0x1; |
| 260 | index = i >> 1; |
| 261 | |
| 262 | /* Update the slice_mask */ |
| 263 | old_psize = (lpsizes[index] >> (mask_index * 4)) & 0xf; |
| 264 | old_mask = slice_mask_for_size(mm, old_psize); |
| 265 | old_mask->low_slices &= ~(1u << i); |
| 266 | psize_mask->low_slices |= 1u << i; |
| 267 | |
| 268 | /* Update the sizes array */ |
| 269 | lpsizes[index] = (lpsizes[index] & ~(0xf << (mask_index * 4))) | |
| 270 | (((unsigned long)psize) << (mask_index * 4)); |
| 271 | } |
| 272 | |
| 273 | hpsizes = mm->context.high_slices_psize; |
| 274 | for (i = 0; i < GET_HIGH_SLICE_INDEX(mm->context.slb_addr_limit); i++) { |
| 275 | if (!test_bit(i, mask->high_slices)) |
| 276 | continue; |
| 277 | |
| 278 | mask_index = i & 0x1; |
| 279 | index = i >> 1; |
| 280 | |
| 281 | /* Update the slice_mask */ |
| 282 | old_psize = (hpsizes[index] >> (mask_index * 4)) & 0xf; |
| 283 | old_mask = slice_mask_for_size(mm, old_psize); |
| 284 | __clear_bit(i, old_mask->high_slices); |
| 285 | __set_bit(i, psize_mask->high_slices); |
| 286 | |
| 287 | /* Update the sizes array */ |
| 288 | hpsizes[index] = (hpsizes[index] & ~(0xf << (mask_index * 4))) | |
| 289 | (((unsigned long)psize) << (mask_index * 4)); |
| 290 | } |
| 291 | |
| 292 | slice_dbg(" lsps=%lx, hsps=%lx\n", |
| 293 | (unsigned long)mm->context.low_slices_psize, |
| 294 | (unsigned long)mm->context.high_slices_psize); |
| 295 | |
| 296 | spin_unlock_irqrestore(&slice_convert_lock, flags); |
| 297 | |
| 298 | copro_flush_all_slbs(mm); |
| 299 | } |
| 300 | |
| 301 | /* |
| 302 | * Compute which slice addr is part of; |
| 303 | * set *boundary_addr to the start or end boundary of that slice |
| 304 | * (depending on 'end' parameter); |
| 305 | * return boolean indicating if the slice is marked as available in the |
| 306 | * 'available' slice_mark. |
| 307 | */ |
| 308 | static bool slice_scan_available(unsigned long addr, |
| 309 | const struct slice_mask *available, |
| 310 | int end, unsigned long *boundary_addr) |
| 311 | { |
| 312 | unsigned long slice; |
| 313 | if (slice_addr_is_low(addr)) { |
| 314 | slice = GET_LOW_SLICE_INDEX(addr); |
| 315 | *boundary_addr = (slice + end) << SLICE_LOW_SHIFT; |
| 316 | return !!(available->low_slices & (1u << slice)); |
| 317 | } else { |
| 318 | slice = GET_HIGH_SLICE_INDEX(addr); |
| 319 | *boundary_addr = (slice + end) ? |
| 320 | ((slice + end) << SLICE_HIGH_SHIFT) : SLICE_LOW_TOP; |
| 321 | return !!test_bit(slice, available->high_slices); |
| 322 | } |
| 323 | } |
| 324 | |
| 325 | static unsigned long slice_find_area_bottomup(struct mm_struct *mm, |
| 326 | unsigned long len, |
| 327 | const struct slice_mask *available, |
| 328 | int psize, unsigned long high_limit) |
| 329 | { |
| 330 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); |
| 331 | unsigned long addr, found, next_end; |
| 332 | struct vm_unmapped_area_info info; |
| 333 | |
| 334 | info.flags = 0; |
| 335 | info.length = len; |
| 336 | info.align_mask = PAGE_MASK & ((1ul << pshift) - 1); |
| 337 | info.align_offset = 0; |
| 338 | |
| 339 | addr = TASK_UNMAPPED_BASE; |
| 340 | /* |
| 341 | * Check till the allow max value for this mmap request |
| 342 | */ |
| 343 | while (addr < high_limit) { |
| 344 | info.low_limit = addr; |
| 345 | if (!slice_scan_available(addr, available, 1, &addr)) |
| 346 | continue; |
| 347 | |
| 348 | next_slice: |
| 349 | /* |
| 350 | * At this point [info.low_limit; addr) covers |
| 351 | * available slices only and ends at a slice boundary. |
| 352 | * Check if we need to reduce the range, or if we can |
| 353 | * extend it to cover the next available slice. |
| 354 | */ |
| 355 | if (addr >= high_limit) |
| 356 | addr = high_limit; |
| 357 | else if (slice_scan_available(addr, available, 1, &next_end)) { |
| 358 | addr = next_end; |
| 359 | goto next_slice; |
| 360 | } |
| 361 | info.high_limit = addr; |
| 362 | |
| 363 | found = vm_unmapped_area(&info); |
| 364 | if (!(found & ~PAGE_MASK)) |
| 365 | return found; |
| 366 | } |
| 367 | |
| 368 | return -ENOMEM; |
| 369 | } |
| 370 | |
| 371 | static unsigned long slice_find_area_topdown(struct mm_struct *mm, |
| 372 | unsigned long len, |
| 373 | const struct slice_mask *available, |
| 374 | int psize, unsigned long high_limit) |
| 375 | { |
| 376 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); |
| 377 | unsigned long addr, found, prev; |
| 378 | struct vm_unmapped_area_info info; |
| 379 | |
| 380 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; |
| 381 | info.length = len; |
| 382 | info.align_mask = PAGE_MASK & ((1ul << pshift) - 1); |
| 383 | info.align_offset = 0; |
| 384 | |
| 385 | addr = mm->mmap_base; |
| 386 | /* |
| 387 | * If we are trying to allocate above DEFAULT_MAP_WINDOW |
| 388 | * Add the different to the mmap_base. |
| 389 | * Only for that request for which high_limit is above |
| 390 | * DEFAULT_MAP_WINDOW we should apply this. |
| 391 | */ |
| 392 | if (high_limit > DEFAULT_MAP_WINDOW) |
| 393 | addr += mm->context.slb_addr_limit - DEFAULT_MAP_WINDOW; |
| 394 | |
| 395 | while (addr > PAGE_SIZE) { |
| 396 | info.high_limit = addr; |
| 397 | if (!slice_scan_available(addr - 1, available, 0, &addr)) |
| 398 | continue; |
| 399 | |
| 400 | prev_slice: |
| 401 | /* |
| 402 | * At this point [addr; info.high_limit) covers |
| 403 | * available slices only and starts at a slice boundary. |
| 404 | * Check if we need to reduce the range, or if we can |
| 405 | * extend it to cover the previous available slice. |
| 406 | */ |
| 407 | if (addr < PAGE_SIZE) |
| 408 | addr = PAGE_SIZE; |
| 409 | else if (slice_scan_available(addr - 1, available, 0, &prev)) { |
| 410 | addr = prev; |
| 411 | goto prev_slice; |
| 412 | } |
| 413 | info.low_limit = addr; |
| 414 | |
| 415 | found = vm_unmapped_area(&info); |
| 416 | if (!(found & ~PAGE_MASK)) |
| 417 | return found; |
| 418 | } |
| 419 | |
| 420 | /* |
| 421 | * A failed mmap() very likely causes application failure, |
| 422 | * so fall back to the bottom-up function here. This scenario |
| 423 | * can happen with large stack limits and large mmap() |
| 424 | * allocations. |
| 425 | */ |
| 426 | return slice_find_area_bottomup(mm, len, available, psize, high_limit); |
| 427 | } |
| 428 | |
| 429 | |
| 430 | static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len, |
| 431 | const struct slice_mask *mask, int psize, |
| 432 | int topdown, unsigned long high_limit) |
| 433 | { |
| 434 | if (topdown) |
| 435 | return slice_find_area_topdown(mm, len, mask, psize, high_limit); |
| 436 | else |
| 437 | return slice_find_area_bottomup(mm, len, mask, psize, high_limit); |
| 438 | } |
| 439 | |
| 440 | static inline void slice_copy_mask(struct slice_mask *dst, |
| 441 | const struct slice_mask *src) |
| 442 | { |
| 443 | dst->low_slices = src->low_slices; |
| 444 | if (!SLICE_NUM_HIGH) |
| 445 | return; |
| 446 | bitmap_copy(dst->high_slices, src->high_slices, SLICE_NUM_HIGH); |
| 447 | } |
| 448 | |
| 449 | static inline void slice_or_mask(struct slice_mask *dst, |
| 450 | const struct slice_mask *src1, |
| 451 | const struct slice_mask *src2) |
| 452 | { |
| 453 | dst->low_slices = src1->low_slices | src2->low_slices; |
| 454 | if (!SLICE_NUM_HIGH) |
| 455 | return; |
| 456 | bitmap_or(dst->high_slices, src1->high_slices, src2->high_slices, SLICE_NUM_HIGH); |
| 457 | } |
| 458 | |
| 459 | static inline void slice_andnot_mask(struct slice_mask *dst, |
| 460 | const struct slice_mask *src1, |
| 461 | const struct slice_mask *src2) |
| 462 | { |
| 463 | dst->low_slices = src1->low_slices & ~src2->low_slices; |
| 464 | if (!SLICE_NUM_HIGH) |
| 465 | return; |
| 466 | bitmap_andnot(dst->high_slices, src1->high_slices, src2->high_slices, SLICE_NUM_HIGH); |
| 467 | } |
| 468 | |
| 469 | #ifdef CONFIG_PPC_64K_PAGES |
| 470 | #define MMU_PAGE_BASE MMU_PAGE_64K |
| 471 | #else |
| 472 | #define MMU_PAGE_BASE MMU_PAGE_4K |
| 473 | #endif |
| 474 | |
| 475 | unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len, |
| 476 | unsigned long flags, unsigned int psize, |
| 477 | int topdown) |
| 478 | { |
| 479 | struct slice_mask good_mask; |
| 480 | struct slice_mask potential_mask; |
| 481 | const struct slice_mask *maskp; |
| 482 | const struct slice_mask *compat_maskp = NULL; |
| 483 | int fixed = (flags & MAP_FIXED); |
| 484 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); |
| 485 | unsigned long page_size = 1UL << pshift; |
| 486 | struct mm_struct *mm = current->mm; |
| 487 | unsigned long newaddr; |
| 488 | unsigned long high_limit; |
| 489 | |
| 490 | high_limit = DEFAULT_MAP_WINDOW; |
| 491 | if (addr >= high_limit || (fixed && (addr + len > high_limit))) |
| 492 | high_limit = TASK_SIZE; |
| 493 | |
| 494 | if (len > high_limit) |
| 495 | return -ENOMEM; |
| 496 | if (len & (page_size - 1)) |
| 497 | return -EINVAL; |
| 498 | if (fixed) { |
| 499 | if (addr & (page_size - 1)) |
| 500 | return -EINVAL; |
| 501 | if (addr > high_limit - len) |
| 502 | return -ENOMEM; |
| 503 | } |
| 504 | |
| 505 | if (high_limit > mm->context.slb_addr_limit) { |
| 506 | /* |
| 507 | * Increasing the slb_addr_limit does not require |
| 508 | * slice mask cache to be recalculated because it should |
| 509 | * be already initialised beyond the old address limit. |
| 510 | */ |
| 511 | mm->context.slb_addr_limit = high_limit; |
| 512 | |
| 513 | on_each_cpu(slice_flush_segments, mm, 1); |
| 514 | } |
| 515 | |
| 516 | /* Sanity checks */ |
| 517 | BUG_ON(mm->task_size == 0); |
| 518 | BUG_ON(mm->context.slb_addr_limit == 0); |
| 519 | VM_BUG_ON(radix_enabled()); |
| 520 | |
| 521 | slice_dbg("slice_get_unmapped_area(mm=%p, psize=%d...\n", mm, psize); |
| 522 | slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d\n", |
| 523 | addr, len, flags, topdown); |
| 524 | |
| 525 | /* If hint, make sure it matches our alignment restrictions */ |
| 526 | if (!fixed && addr) { |
| 527 | addr = _ALIGN_UP(addr, page_size); |
| 528 | slice_dbg(" aligned addr=%lx\n", addr); |
| 529 | /* Ignore hint if it's too large or overlaps a VMA */ |
| 530 | if (addr > high_limit - len || |
| 531 | !slice_area_is_free(mm, addr, len)) |
| 532 | addr = 0; |
| 533 | } |
| 534 | |
| 535 | /* First make up a "good" mask of slices that have the right size |
| 536 | * already |
| 537 | */ |
| 538 | maskp = slice_mask_for_size(mm, psize); |
| 539 | |
| 540 | /* |
| 541 | * Here "good" means slices that are already the right page size, |
| 542 | * "compat" means slices that have a compatible page size (i.e. |
| 543 | * 4k in a 64k pagesize kernel), and "free" means slices without |
| 544 | * any VMAs. |
| 545 | * |
| 546 | * If MAP_FIXED: |
| 547 | * check if fits in good | compat => OK |
| 548 | * check if fits in good | compat | free => convert free |
| 549 | * else bad |
| 550 | * If have hint: |
| 551 | * check if hint fits in good => OK |
| 552 | * check if hint fits in good | free => convert free |
| 553 | * Otherwise: |
| 554 | * search in good, found => OK |
| 555 | * search in good | free, found => convert free |
| 556 | * search in good | compat | free, found => convert free. |
| 557 | */ |
| 558 | |
| 559 | /* |
| 560 | * If we support combo pages, we can allow 64k pages in 4k slices |
| 561 | * The mask copies could be avoided in most cases here if we had |
| 562 | * a pointer to good mask for the next code to use. |
| 563 | */ |
| 564 | if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && psize == MMU_PAGE_64K) { |
| 565 | compat_maskp = slice_mask_for_size(mm, MMU_PAGE_4K); |
| 566 | if (fixed) |
| 567 | slice_or_mask(&good_mask, maskp, compat_maskp); |
| 568 | else |
| 569 | slice_copy_mask(&good_mask, maskp); |
| 570 | } else { |
| 571 | slice_copy_mask(&good_mask, maskp); |
| 572 | } |
| 573 | |
| 574 | slice_print_mask(" good_mask", &good_mask); |
| 575 | if (compat_maskp) |
| 576 | slice_print_mask(" compat_mask", compat_maskp); |
| 577 | |
| 578 | /* First check hint if it's valid or if we have MAP_FIXED */ |
| 579 | if (addr != 0 || fixed) { |
| 580 | /* Check if we fit in the good mask. If we do, we just return, |
| 581 | * nothing else to do |
| 582 | */ |
| 583 | if (slice_check_range_fits(mm, &good_mask, addr, len)) { |
| 584 | slice_dbg(" fits good !\n"); |
| 585 | newaddr = addr; |
| 586 | goto return_addr; |
| 587 | } |
| 588 | } else { |
| 589 | /* Now let's see if we can find something in the existing |
| 590 | * slices for that size |
| 591 | */ |
| 592 | newaddr = slice_find_area(mm, len, &good_mask, |
| 593 | psize, topdown, high_limit); |
| 594 | if (newaddr != -ENOMEM) { |
| 595 | /* Found within the good mask, we don't have to setup, |
| 596 | * we thus return directly |
| 597 | */ |
| 598 | slice_dbg(" found area at 0x%lx\n", newaddr); |
| 599 | goto return_addr; |
| 600 | } |
| 601 | } |
| 602 | /* |
| 603 | * We don't fit in the good mask, check what other slices are |
| 604 | * empty and thus can be converted |
| 605 | */ |
| 606 | slice_mask_for_free(mm, &potential_mask, high_limit); |
| 607 | slice_or_mask(&potential_mask, &potential_mask, &good_mask); |
| 608 | slice_print_mask(" potential", &potential_mask); |
| 609 | |
| 610 | if (addr != 0 || fixed) { |
| 611 | if (slice_check_range_fits(mm, &potential_mask, addr, len)) { |
| 612 | slice_dbg(" fits potential !\n"); |
| 613 | newaddr = addr; |
| 614 | goto convert; |
| 615 | } |
| 616 | } |
| 617 | |
| 618 | /* If we have MAP_FIXED and failed the above steps, then error out */ |
| 619 | if (fixed) |
| 620 | return -EBUSY; |
| 621 | |
| 622 | slice_dbg(" search...\n"); |
| 623 | |
| 624 | /* If we had a hint that didn't work out, see if we can fit |
| 625 | * anywhere in the good area. |
| 626 | */ |
| 627 | if (addr) { |
| 628 | newaddr = slice_find_area(mm, len, &good_mask, |
| 629 | psize, topdown, high_limit); |
| 630 | if (newaddr != -ENOMEM) { |
| 631 | slice_dbg(" found area at 0x%lx\n", newaddr); |
| 632 | goto return_addr; |
| 633 | } |
| 634 | } |
| 635 | |
| 636 | /* Now let's see if we can find something in the existing slices |
| 637 | * for that size plus free slices |
| 638 | */ |
| 639 | newaddr = slice_find_area(mm, len, &potential_mask, |
| 640 | psize, topdown, high_limit); |
| 641 | |
| 642 | #ifdef CONFIG_PPC_64K_PAGES |
| 643 | if (newaddr == -ENOMEM && psize == MMU_PAGE_64K) { |
| 644 | /* retry the search with 4k-page slices included */ |
| 645 | slice_or_mask(&potential_mask, &potential_mask, compat_maskp); |
| 646 | newaddr = slice_find_area(mm, len, &potential_mask, |
| 647 | psize, topdown, high_limit); |
| 648 | } |
| 649 | #endif |
| 650 | |
| 651 | if (newaddr == -ENOMEM) |
| 652 | return -ENOMEM; |
| 653 | |
| 654 | slice_range_to_mask(newaddr, len, &potential_mask); |
| 655 | slice_dbg(" found potential area at 0x%lx\n", newaddr); |
| 656 | slice_print_mask(" mask", &potential_mask); |
| 657 | |
| 658 | convert: |
| 659 | /* |
| 660 | * Try to allocate the context before we do slice convert |
| 661 | * so that we handle the context allocation failure gracefully. |
| 662 | */ |
| 663 | if (need_extra_context(mm, newaddr)) { |
| 664 | if (alloc_extended_context(mm, newaddr) < 0) |
| 665 | return -ENOMEM; |
| 666 | } |
| 667 | |
| 668 | slice_andnot_mask(&potential_mask, &potential_mask, &good_mask); |
| 669 | if (compat_maskp && !fixed) |
| 670 | slice_andnot_mask(&potential_mask, &potential_mask, compat_maskp); |
| 671 | if (potential_mask.low_slices || |
| 672 | (SLICE_NUM_HIGH && |
| 673 | !bitmap_empty(potential_mask.high_slices, SLICE_NUM_HIGH))) { |
| 674 | slice_convert(mm, &potential_mask, psize); |
| 675 | if (psize > MMU_PAGE_BASE) |
| 676 | on_each_cpu(slice_flush_segments, mm, 1); |
| 677 | } |
| 678 | return newaddr; |
| 679 | |
| 680 | return_addr: |
| 681 | if (need_extra_context(mm, newaddr)) { |
| 682 | if (alloc_extended_context(mm, newaddr) < 0) |
| 683 | return -ENOMEM; |
| 684 | } |
| 685 | return newaddr; |
| 686 | } |
| 687 | EXPORT_SYMBOL_GPL(slice_get_unmapped_area); |
| 688 | |
| 689 | unsigned long arch_get_unmapped_area(struct file *filp, |
| 690 | unsigned long addr, |
| 691 | unsigned long len, |
| 692 | unsigned long pgoff, |
| 693 | unsigned long flags) |
| 694 | { |
| 695 | return slice_get_unmapped_area(addr, len, flags, |
| 696 | current->mm->context.user_psize, 0); |
| 697 | } |
| 698 | |
| 699 | unsigned long arch_get_unmapped_area_topdown(struct file *filp, |
| 700 | const unsigned long addr0, |
| 701 | const unsigned long len, |
| 702 | const unsigned long pgoff, |
| 703 | const unsigned long flags) |
| 704 | { |
| 705 | return slice_get_unmapped_area(addr0, len, flags, |
| 706 | current->mm->context.user_psize, 1); |
| 707 | } |
| 708 | |
| 709 | unsigned int get_slice_psize(struct mm_struct *mm, unsigned long addr) |
| 710 | { |
| 711 | unsigned char *psizes; |
| 712 | int index, mask_index; |
| 713 | |
| 714 | VM_BUG_ON(radix_enabled()); |
| 715 | |
| 716 | if (slice_addr_is_low(addr)) { |
| 717 | psizes = mm->context.low_slices_psize; |
| 718 | index = GET_LOW_SLICE_INDEX(addr); |
| 719 | } else { |
| 720 | psizes = mm->context.high_slices_psize; |
| 721 | index = GET_HIGH_SLICE_INDEX(addr); |
| 722 | } |
| 723 | mask_index = index & 0x1; |
| 724 | return (psizes[index >> 1] >> (mask_index * 4)) & 0xf; |
| 725 | } |
| 726 | EXPORT_SYMBOL_GPL(get_slice_psize); |
| 727 | |
| 728 | void slice_init_new_context_exec(struct mm_struct *mm) |
| 729 | { |
| 730 | unsigned char *hpsizes, *lpsizes; |
| 731 | struct slice_mask *mask; |
| 732 | unsigned int psize = mmu_virtual_psize; |
| 733 | |
| 734 | slice_dbg("slice_init_new_context_exec(mm=%p)\n", mm); |
| 735 | |
| 736 | /* |
| 737 | * In the case of exec, use the default limit. In the |
| 738 | * case of fork it is just inherited from the mm being |
| 739 | * duplicated. |
| 740 | */ |
| 741 | #ifdef CONFIG_PPC64 |
| 742 | mm->context.slb_addr_limit = DEFAULT_MAP_WINDOW_USER64; |
| 743 | #else |
| 744 | mm->context.slb_addr_limit = DEFAULT_MAP_WINDOW; |
| 745 | #endif |
| 746 | |
| 747 | mm->context.user_psize = psize; |
| 748 | |
| 749 | /* |
| 750 | * Set all slice psizes to the default. |
| 751 | */ |
| 752 | lpsizes = mm->context.low_slices_psize; |
| 753 | memset(lpsizes, (psize << 4) | psize, SLICE_NUM_LOW >> 1); |
| 754 | |
| 755 | hpsizes = mm->context.high_slices_psize; |
| 756 | memset(hpsizes, (psize << 4) | psize, SLICE_NUM_HIGH >> 1); |
| 757 | |
| 758 | /* |
| 759 | * Slice mask cache starts zeroed, fill the default size cache. |
| 760 | */ |
| 761 | mask = slice_mask_for_size(mm, psize); |
| 762 | mask->low_slices = ~0UL; |
| 763 | if (SLICE_NUM_HIGH) |
| 764 | bitmap_fill(mask->high_slices, SLICE_NUM_HIGH); |
| 765 | } |
| 766 | |
| 767 | void slice_set_range_psize(struct mm_struct *mm, unsigned long start, |
| 768 | unsigned long len, unsigned int psize) |
| 769 | { |
| 770 | struct slice_mask mask; |
| 771 | |
| 772 | VM_BUG_ON(radix_enabled()); |
| 773 | |
| 774 | slice_range_to_mask(start, len, &mask); |
| 775 | slice_convert(mm, &mask, psize); |
| 776 | } |
| 777 | |
| 778 | #ifdef CONFIG_HUGETLB_PAGE |
| 779 | /* |
| 780 | * is_hugepage_only_range() is used by generic code to verify whether |
| 781 | * a normal mmap mapping (non hugetlbfs) is valid on a given area. |
| 782 | * |
| 783 | * until the generic code provides a more generic hook and/or starts |
| 784 | * calling arch get_unmapped_area for MAP_FIXED (which our implementation |
| 785 | * here knows how to deal with), we hijack it to keep standard mappings |
| 786 | * away from us. |
| 787 | * |
| 788 | * because of that generic code limitation, MAP_FIXED mapping cannot |
| 789 | * "convert" back a slice with no VMAs to the standard page size, only |
| 790 | * get_unmapped_area() can. It would be possible to fix it here but I |
| 791 | * prefer working on fixing the generic code instead. |
| 792 | * |
| 793 | * WARNING: This will not work if hugetlbfs isn't enabled since the |
| 794 | * generic code will redefine that function as 0 in that. This is ok |
| 795 | * for now as we only use slices with hugetlbfs enabled. This should |
| 796 | * be fixed as the generic code gets fixed. |
| 797 | */ |
| 798 | int slice_is_hugepage_only_range(struct mm_struct *mm, unsigned long addr, |
| 799 | unsigned long len) |
| 800 | { |
| 801 | const struct slice_mask *maskp; |
| 802 | unsigned int psize = mm->context.user_psize; |
| 803 | |
| 804 | VM_BUG_ON(radix_enabled()); |
| 805 | |
| 806 | maskp = slice_mask_for_size(mm, psize); |
| 807 | #ifdef CONFIG_PPC_64K_PAGES |
| 808 | /* We need to account for 4k slices too */ |
| 809 | if (psize == MMU_PAGE_64K) { |
| 810 | const struct slice_mask *compat_maskp; |
| 811 | struct slice_mask available; |
| 812 | |
| 813 | compat_maskp = slice_mask_for_size(mm, MMU_PAGE_4K); |
| 814 | slice_or_mask(&available, maskp, compat_maskp); |
| 815 | return !slice_check_range_fits(mm, &available, addr, len); |
| 816 | } |
| 817 | #endif |
| 818 | |
| 819 | return !slice_check_range_fits(mm, maskp, addr, len); |
| 820 | } |
| 821 | #endif |