Commit | Line | Data |
---|---|---|
d0f13e3c BH |
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> | |
4b16f8e2 | 32 | #include <linux/export.h> |
1217d34b | 33 | #include <linux/hugetlb.h> |
d0f13e3c BH |
34 | #include <asm/mman.h> |
35 | #include <asm/mmu.h> | |
be3ebfe8 | 36 | #include <asm/copro.h> |
1217d34b | 37 | #include <asm/hugetlb.h> |
d0f13e3c | 38 | |
f7a75f0a | 39 | static DEFINE_SPINLOCK(slice_convert_lock); |
82185222 AK |
40 | /* |
41 | * One bit per slice. We have lower slices which cover 256MB segments | |
42 | * upto 4G range. That gets us 16 low slices. For the rest we track slices | |
43 | * in 1TB size. | |
44 | */ | |
45 | struct slice_mask { | |
46 | u64 low_slices; | |
47 | DECLARE_BITMAP(high_slices, SLICE_NUM_HIGH); | |
48 | }; | |
d0f13e3c BH |
49 | |
50 | #ifdef DEBUG | |
51 | int _slice_debug = 1; | |
52 | ||
53 | static void slice_print_mask(const char *label, struct slice_mask mask) | |
54 | { | |
d0f13e3c BH |
55 | if (!_slice_debug) |
56 | return; | |
302413ca AK |
57 | pr_devel("%s low_slice: %*pbl\n", label, (int)SLICE_NUM_LOW, &mask.low_slices); |
58 | pr_devel("%s high_slice: %*pbl\n", label, (int)SLICE_NUM_HIGH, mask.high_slices); | |
d0f13e3c BH |
59 | } |
60 | ||
302413ca | 61 | #define slice_dbg(fmt...) do { if (_slice_debug) pr_devel(fmt); } while (0) |
d0f13e3c BH |
62 | |
63 | #else | |
64 | ||
65 | static void slice_print_mask(const char *label, struct slice_mask mask) {} | |
66 | #define slice_dbg(fmt...) | |
67 | ||
68 | #endif | |
69 | ||
a4d36215 AK |
70 | static void slice_range_to_mask(unsigned long start, unsigned long len, |
71 | struct slice_mask *ret) | |
d0f13e3c BH |
72 | { |
73 | unsigned long end = start + len - 1; | |
f3207c12 | 74 | |
a4d36215 AK |
75 | ret->low_slices = 0; |
76 | bitmap_zero(ret->high_slices, SLICE_NUM_HIGH); | |
d0f13e3c BH |
77 | |
78 | if (start < SLICE_LOW_TOP) { | |
98beda74 | 79 | unsigned long mend = min(end, (SLICE_LOW_TOP - 1)); |
d0f13e3c | 80 | |
a4d36215 | 81 | ret->low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1)) |
98beda74 | 82 | - (1u << GET_LOW_SLICE_INDEX(start)); |
d0f13e3c BH |
83 | } |
84 | ||
f3207c12 AK |
85 | if ((start + len) > SLICE_LOW_TOP) { |
86 | unsigned long start_index = GET_HIGH_SLICE_INDEX(start); | |
87 | unsigned long align_end = ALIGN(end, (1UL << SLICE_HIGH_SHIFT)); | |
88 | unsigned long count = GET_HIGH_SLICE_INDEX(align_end) - start_index; | |
d0f13e3c | 89 | |
a4d36215 | 90 | bitmap_set(ret->high_slices, start_index, count); |
f3207c12 | 91 | } |
d0f13e3c BH |
92 | } |
93 | ||
94 | static int slice_area_is_free(struct mm_struct *mm, unsigned long addr, | |
95 | unsigned long len) | |
96 | { | |
97 | struct vm_area_struct *vma; | |
98 | ||
4722476b | 99 | if ((mm->context.slb_addr_limit - len) < addr) |
d0f13e3c BH |
100 | return 0; |
101 | vma = find_vma(mm, addr); | |
1be7107f | 102 | return (!vma || (addr + len) <= vm_start_gap(vma)); |
d0f13e3c BH |
103 | } |
104 | ||
105 | static int slice_low_has_vma(struct mm_struct *mm, unsigned long slice) | |
106 | { | |
107 | return !slice_area_is_free(mm, slice << SLICE_LOW_SHIFT, | |
108 | 1ul << SLICE_LOW_SHIFT); | |
109 | } | |
110 | ||
111 | static int slice_high_has_vma(struct mm_struct *mm, unsigned long slice) | |
112 | { | |
113 | unsigned long start = slice << SLICE_HIGH_SHIFT; | |
114 | unsigned long end = start + (1ul << SLICE_HIGH_SHIFT); | |
115 | ||
116 | /* Hack, so that each addresses is controlled by exactly one | |
117 | * of the high or low area bitmaps, the first high area starts | |
118 | * at 4GB, not 0 */ | |
119 | if (start == 0) | |
120 | start = SLICE_LOW_TOP; | |
121 | ||
122 | return !slice_area_is_free(mm, start, end - start); | |
123 | } | |
124 | ||
7a06c668 AK |
125 | static void slice_mask_for_free(struct mm_struct *mm, struct slice_mask *ret, |
126 | unsigned long high_limit) | |
d0f13e3c | 127 | { |
d0f13e3c BH |
128 | unsigned long i; |
129 | ||
a4d36215 AK |
130 | ret->low_slices = 0; |
131 | bitmap_zero(ret->high_slices, SLICE_NUM_HIGH); | |
f3207c12 | 132 | |
d0f13e3c BH |
133 | for (i = 0; i < SLICE_NUM_LOW; i++) |
134 | if (!slice_low_has_vma(mm, i)) | |
a4d36215 | 135 | ret->low_slices |= 1u << i; |
d0f13e3c | 136 | |
7a06c668 | 137 | if (high_limit <= SLICE_LOW_TOP) |
a4d36215 | 138 | return; |
d0f13e3c | 139 | |
7a06c668 | 140 | for (i = 0; i < GET_HIGH_SLICE_INDEX(high_limit); i++) |
d0f13e3c | 141 | if (!slice_high_has_vma(mm, i)) |
a4d36215 | 142 | __set_bit(i, ret->high_slices); |
d0f13e3c BH |
143 | } |
144 | ||
7a06c668 AK |
145 | static void slice_mask_for_size(struct mm_struct *mm, int psize, struct slice_mask *ret, |
146 | unsigned long high_limit) | |
d0f13e3c | 147 | { |
7aa0727f AK |
148 | unsigned char *hpsizes; |
149 | int index, mask_index; | |
d0f13e3c | 150 | unsigned long i; |
7aa0727f | 151 | u64 lpsizes; |
d0f13e3c | 152 | |
a4d36215 AK |
153 | ret->low_slices = 0; |
154 | bitmap_zero(ret->high_slices, SLICE_NUM_HIGH); | |
f3207c12 | 155 | |
7aa0727f | 156 | lpsizes = mm->context.low_slices_psize; |
d0f13e3c | 157 | for (i = 0; i < SLICE_NUM_LOW; i++) |
7aa0727f | 158 | if (((lpsizes >> (i * 4)) & 0xf) == psize) |
a4d36215 | 159 | ret->low_slices |= 1u << i; |
d0f13e3c | 160 | |
7a06c668 AK |
161 | if (high_limit <= SLICE_LOW_TOP) |
162 | return; | |
163 | ||
7aa0727f | 164 | hpsizes = mm->context.high_slices_psize; |
7a06c668 | 165 | for (i = 0; i < GET_HIGH_SLICE_INDEX(high_limit); i++) { |
7aa0727f AK |
166 | mask_index = i & 0x1; |
167 | index = i >> 1; | |
168 | if (((hpsizes[index] >> (mask_index * 4)) & 0xf) == psize) | |
a4d36215 | 169 | __set_bit(i, ret->high_slices); |
7aa0727f | 170 | } |
d0f13e3c BH |
171 | } |
172 | ||
957b778a AK |
173 | static int slice_check_fit(struct mm_struct *mm, |
174 | struct slice_mask mask, struct slice_mask available) | |
d0f13e3c | 175 | { |
f3207c12 | 176 | DECLARE_BITMAP(result, SLICE_NUM_HIGH); |
7a06c668 AK |
177 | /* |
178 | * Make sure we just do bit compare only to the max | |
179 | * addr limit and not the full bit map size. | |
180 | */ | |
4722476b | 181 | unsigned long slice_count = GET_HIGH_SLICE_INDEX(mm->context.slb_addr_limit); |
f3207c12 AK |
182 | |
183 | bitmap_and(result, mask.high_slices, | |
957b778a | 184 | available.high_slices, slice_count); |
f3207c12 | 185 | |
d0f13e3c | 186 | return (mask.low_slices & available.low_slices) == mask.low_slices && |
957b778a | 187 | bitmap_equal(result, mask.high_slices, slice_count); |
d0f13e3c BH |
188 | } |
189 | ||
190 | static void slice_flush_segments(void *parm) | |
191 | { | |
192 | struct mm_struct *mm = parm; | |
193 | unsigned long flags; | |
194 | ||
195 | if (mm != current->active_mm) | |
196 | return; | |
197 | ||
52b1e665 | 198 | copy_mm_to_paca(current->active_mm); |
d0f13e3c BH |
199 | |
200 | local_irq_save(flags); | |
201 | slb_flush_and_rebolt(); | |
202 | local_irq_restore(flags); | |
203 | } | |
204 | ||
205 | static void slice_convert(struct mm_struct *mm, struct slice_mask mask, int psize) | |
206 | { | |
7aa0727f | 207 | int index, mask_index; |
d0f13e3c | 208 | /* Write the new slice psize bits */ |
7aa0727f AK |
209 | unsigned char *hpsizes; |
210 | u64 lpsizes; | |
d0f13e3c BH |
211 | unsigned long i, flags; |
212 | ||
213 | slice_dbg("slice_convert(mm=%p, psize=%d)\n", mm, psize); | |
214 | slice_print_mask(" mask", mask); | |
215 | ||
216 | /* We need to use a spinlock here to protect against | |
217 | * concurrent 64k -> 4k demotion ... | |
218 | */ | |
219 | spin_lock_irqsave(&slice_convert_lock, flags); | |
220 | ||
221 | lpsizes = mm->context.low_slices_psize; | |
222 | for (i = 0; i < SLICE_NUM_LOW; i++) | |
223 | if (mask.low_slices & (1u << i)) | |
224 | lpsizes = (lpsizes & ~(0xful << (i * 4))) | | |
225 | (((unsigned long)psize) << (i * 4)); | |
226 | ||
7aa0727f | 227 | /* Assign the value back */ |
d0f13e3c | 228 | mm->context.low_slices_psize = lpsizes; |
7aa0727f AK |
229 | |
230 | hpsizes = mm->context.high_slices_psize; | |
4722476b | 231 | for (i = 0; i < GET_HIGH_SLICE_INDEX(mm->context.slb_addr_limit); i++) { |
7aa0727f AK |
232 | mask_index = i & 0x1; |
233 | index = i >> 1; | |
f3207c12 | 234 | if (test_bit(i, mask.high_slices)) |
7aa0727f AK |
235 | hpsizes[index] = (hpsizes[index] & |
236 | ~(0xf << (mask_index * 4))) | | |
237 | (((unsigned long)psize) << (mask_index * 4)); | |
238 | } | |
d0f13e3c BH |
239 | |
240 | slice_dbg(" lsps=%lx, hsps=%lx\n", | |
302413ca AK |
241 | (unsigned long)mm->context.low_slices_psize, |
242 | (unsigned long)mm->context.high_slices_psize); | |
d0f13e3c BH |
243 | |
244 | spin_unlock_irqrestore(&slice_convert_lock, flags); | |
d0f13e3c | 245 | |
be3ebfe8 | 246 | copro_flush_all_slbs(mm); |
d0f13e3c BH |
247 | } |
248 | ||
fba2369e ML |
249 | /* |
250 | * Compute which slice addr is part of; | |
251 | * set *boundary_addr to the start or end boundary of that slice | |
252 | * (depending on 'end' parameter); | |
253 | * return boolean indicating if the slice is marked as available in the | |
254 | * 'available' slice_mark. | |
255 | */ | |
256 | static bool slice_scan_available(unsigned long addr, | |
257 | struct slice_mask available, | |
258 | int end, | |
259 | unsigned long *boundary_addr) | |
260 | { | |
261 | unsigned long slice; | |
262 | if (addr < SLICE_LOW_TOP) { | |
263 | slice = GET_LOW_SLICE_INDEX(addr); | |
264 | *boundary_addr = (slice + end) << SLICE_LOW_SHIFT; | |
265 | return !!(available.low_slices & (1u << slice)); | |
266 | } else { | |
267 | slice = GET_HIGH_SLICE_INDEX(addr); | |
268 | *boundary_addr = (slice + end) ? | |
269 | ((slice + end) << SLICE_HIGH_SHIFT) : SLICE_LOW_TOP; | |
f3207c12 | 270 | return !!test_bit(slice, available.high_slices); |
fba2369e ML |
271 | } |
272 | } | |
273 | ||
d0f13e3c BH |
274 | static unsigned long slice_find_area_bottomup(struct mm_struct *mm, |
275 | unsigned long len, | |
276 | struct slice_mask available, | |
f4ea6dcb | 277 | int psize, unsigned long high_limit) |
d0f13e3c | 278 | { |
d0f13e3c | 279 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); |
fba2369e ML |
280 | unsigned long addr, found, next_end; |
281 | struct vm_unmapped_area_info info; | |
d0f13e3c | 282 | |
fba2369e ML |
283 | info.flags = 0; |
284 | info.length = len; | |
285 | info.align_mask = PAGE_MASK & ((1ul << pshift) - 1); | |
286 | info.align_offset = 0; | |
d0f13e3c | 287 | |
fba2369e | 288 | addr = TASK_UNMAPPED_BASE; |
f4ea6dcb AK |
289 | /* |
290 | * Check till the allow max value for this mmap request | |
291 | */ | |
292 | while (addr < high_limit) { | |
fba2369e ML |
293 | info.low_limit = addr; |
294 | if (!slice_scan_available(addr, available, 1, &addr)) | |
d0f13e3c | 295 | continue; |
fba2369e ML |
296 | |
297 | next_slice: | |
298 | /* | |
299 | * At this point [info.low_limit; addr) covers | |
300 | * available slices only and ends at a slice boundary. | |
301 | * Check if we need to reduce the range, or if we can | |
302 | * extend it to cover the next available slice. | |
303 | */ | |
be77e999 AK |
304 | if (addr >= high_limit) |
305 | addr = high_limit; | |
fba2369e ML |
306 | else if (slice_scan_available(addr, available, 1, &next_end)) { |
307 | addr = next_end; | |
308 | goto next_slice; | |
d0f13e3c | 309 | } |
fba2369e ML |
310 | info.high_limit = addr; |
311 | ||
312 | found = vm_unmapped_area(&info); | |
313 | if (!(found & ~PAGE_MASK)) | |
314 | return found; | |
d0f13e3c BH |
315 | } |
316 | ||
d0f13e3c BH |
317 | return -ENOMEM; |
318 | } | |
319 | ||
320 | static unsigned long slice_find_area_topdown(struct mm_struct *mm, | |
321 | unsigned long len, | |
322 | struct slice_mask available, | |
f4ea6dcb | 323 | int psize, unsigned long high_limit) |
d0f13e3c | 324 | { |
d0f13e3c | 325 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); |
fba2369e ML |
326 | unsigned long addr, found, prev; |
327 | struct vm_unmapped_area_info info; | |
328 | ||
329 | info.flags = VM_UNMAPPED_AREA_TOPDOWN; | |
330 | info.length = len; | |
331 | info.align_mask = PAGE_MASK & ((1ul << pshift) - 1); | |
332 | info.align_offset = 0; | |
d0f13e3c | 333 | |
d0f13e3c | 334 | addr = mm->mmap_base; |
f4ea6dcb AK |
335 | /* |
336 | * If we are trying to allocate above DEFAULT_MAP_WINDOW | |
337 | * Add the different to the mmap_base. | |
338 | * Only for that request for which high_limit is above | |
339 | * DEFAULT_MAP_WINDOW we should apply this. | |
340 | */ | |
4722476b NP |
341 | if (high_limit > DEFAULT_MAP_WINDOW) |
342 | addr += mm->context.slb_addr_limit - DEFAULT_MAP_WINDOW; | |
f4ea6dcb | 343 | |
fba2369e ML |
344 | while (addr > PAGE_SIZE) { |
345 | info.high_limit = addr; | |
346 | if (!slice_scan_available(addr - 1, available, 0, &addr)) | |
d0f13e3c | 347 | continue; |
d0f13e3c | 348 | |
fba2369e | 349 | prev_slice: |
d0f13e3c | 350 | /* |
fba2369e ML |
351 | * At this point [addr; info.high_limit) covers |
352 | * available slices only and starts at a slice boundary. | |
353 | * Check if we need to reduce the range, or if we can | |
354 | * extend it to cover the previous available slice. | |
d0f13e3c | 355 | */ |
fba2369e ML |
356 | if (addr < PAGE_SIZE) |
357 | addr = PAGE_SIZE; | |
358 | else if (slice_scan_available(addr - 1, available, 0, &prev)) { | |
359 | addr = prev; | |
360 | goto prev_slice; | |
361 | } | |
362 | info.low_limit = addr; | |
d0f13e3c | 363 | |
fba2369e ML |
364 | found = vm_unmapped_area(&info); |
365 | if (!(found & ~PAGE_MASK)) | |
366 | return found; | |
d0f13e3c BH |
367 | } |
368 | ||
369 | /* | |
370 | * A failed mmap() very likely causes application failure, | |
371 | * so fall back to the bottom-up function here. This scenario | |
372 | * can happen with large stack limits and large mmap() | |
373 | * allocations. | |
374 | */ | |
f4ea6dcb | 375 | return slice_find_area_bottomup(mm, len, available, psize, high_limit); |
d0f13e3c BH |
376 | } |
377 | ||
378 | ||
379 | static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len, | |
380 | struct slice_mask mask, int psize, | |
f4ea6dcb | 381 | int topdown, unsigned long high_limit) |
d0f13e3c BH |
382 | { |
383 | if (topdown) | |
f4ea6dcb | 384 | return slice_find_area_topdown(mm, len, mask, psize, high_limit); |
d0f13e3c | 385 | else |
f4ea6dcb | 386 | return slice_find_area_bottomup(mm, len, mask, psize, high_limit); |
d0f13e3c BH |
387 | } |
388 | ||
f3207c12 AK |
389 | static inline void slice_or_mask(struct slice_mask *dst, struct slice_mask *src) |
390 | { | |
391 | DECLARE_BITMAP(result, SLICE_NUM_HIGH); | |
3a8247cc | 392 | |
f3207c12 AK |
393 | dst->low_slices |= src->low_slices; |
394 | bitmap_or(result, dst->high_slices, src->high_slices, SLICE_NUM_HIGH); | |
395 | bitmap_copy(dst->high_slices, result, SLICE_NUM_HIGH); | |
396 | } | |
397 | ||
398 | static inline void slice_andnot_mask(struct slice_mask *dst, struct slice_mask *src) | |
399 | { | |
400 | DECLARE_BITMAP(result, SLICE_NUM_HIGH); | |
401 | ||
402 | dst->low_slices &= ~src->low_slices; | |
403 | ||
404 | bitmap_andnot(result, dst->high_slices, src->high_slices, SLICE_NUM_HIGH); | |
405 | bitmap_copy(dst->high_slices, result, SLICE_NUM_HIGH); | |
406 | } | |
3a8247cc PM |
407 | |
408 | #ifdef CONFIG_PPC_64K_PAGES | |
409 | #define MMU_PAGE_BASE MMU_PAGE_64K | |
410 | #else | |
411 | #define MMU_PAGE_BASE MMU_PAGE_4K | |
412 | #endif | |
413 | ||
d0f13e3c BH |
414 | unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len, |
415 | unsigned long flags, unsigned int psize, | |
34d07177 | 416 | int topdown) |
d0f13e3c | 417 | { |
f3207c12 | 418 | struct slice_mask mask; |
d0f13e3c | 419 | struct slice_mask good_mask; |
f3207c12 AK |
420 | struct slice_mask potential_mask; |
421 | struct slice_mask compat_mask; | |
d0f13e3c BH |
422 | int fixed = (flags & MAP_FIXED); |
423 | int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT); | |
6a72dc03 | 424 | unsigned long page_size = 1UL << pshift; |
d0f13e3c | 425 | struct mm_struct *mm = current->mm; |
3a8247cc | 426 | unsigned long newaddr; |
f4ea6dcb | 427 | unsigned long high_limit; |
d0f13e3c | 428 | |
6a72dc03 | 429 | high_limit = DEFAULT_MAP_WINDOW; |
35602f82 | 430 | if (addr >= high_limit || (fixed && (addr + len > high_limit))) |
6a72dc03 NP |
431 | high_limit = TASK_SIZE; |
432 | ||
433 | if (len > high_limit) | |
434 | return -ENOMEM; | |
435 | if (len & (page_size - 1)) | |
436 | return -EINVAL; | |
437 | if (fixed) { | |
438 | if (addr & (page_size - 1)) | |
439 | return -EINVAL; | |
440 | if (addr > high_limit - len) | |
441 | return -ENOMEM; | |
442 | } | |
443 | ||
4722476b NP |
444 | if (high_limit > mm->context.slb_addr_limit) { |
445 | mm->context.slb_addr_limit = high_limit; | |
f4ea6dcb AK |
446 | on_each_cpu(slice_flush_segments, mm, 1); |
447 | } | |
6a72dc03 | 448 | |
f3207c12 AK |
449 | /* |
450 | * init different masks | |
451 | */ | |
452 | mask.low_slices = 0; | |
453 | bitmap_zero(mask.high_slices, SLICE_NUM_HIGH); | |
454 | ||
455 | /* silence stupid warning */; | |
456 | potential_mask.low_slices = 0; | |
457 | bitmap_zero(potential_mask.high_slices, SLICE_NUM_HIGH); | |
458 | ||
459 | compat_mask.low_slices = 0; | |
460 | bitmap_zero(compat_mask.high_slices, SLICE_NUM_HIGH); | |
461 | ||
d0f13e3c BH |
462 | /* Sanity checks */ |
463 | BUG_ON(mm->task_size == 0); | |
4722476b | 464 | BUG_ON(mm->context.slb_addr_limit == 0); |
764041e0 | 465 | VM_BUG_ON(radix_enabled()); |
d0f13e3c BH |
466 | |
467 | slice_dbg("slice_get_unmapped_area(mm=%p, psize=%d...\n", mm, psize); | |
34d07177 ML |
468 | slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d\n", |
469 | addr, len, flags, topdown); | |
d0f13e3c | 470 | |
d0f13e3c BH |
471 | /* If hint, make sure it matches our alignment restrictions */ |
472 | if (!fixed && addr) { | |
6a72dc03 | 473 | addr = _ALIGN_UP(addr, page_size); |
d0f13e3c | 474 | slice_dbg(" aligned addr=%lx\n", addr); |
3a8247cc | 475 | /* Ignore hint if it's too large or overlaps a VMA */ |
6a72dc03 | 476 | if (addr > high_limit - len || |
3a8247cc PM |
477 | !slice_area_is_free(mm, addr, len)) |
478 | addr = 0; | |
d0f13e3c BH |
479 | } |
480 | ||
3a8247cc | 481 | /* First make up a "good" mask of slices that have the right size |
d0f13e3c BH |
482 | * already |
483 | */ | |
7a06c668 | 484 | slice_mask_for_size(mm, psize, &good_mask, high_limit); |
d0f13e3c BH |
485 | slice_print_mask(" good_mask", good_mask); |
486 | ||
3a8247cc PM |
487 | /* |
488 | * Here "good" means slices that are already the right page size, | |
489 | * "compat" means slices that have a compatible page size (i.e. | |
490 | * 4k in a 64k pagesize kernel), and "free" means slices without | |
491 | * any VMAs. | |
492 | * | |
493 | * If MAP_FIXED: | |
494 | * check if fits in good | compat => OK | |
495 | * check if fits in good | compat | free => convert free | |
496 | * else bad | |
497 | * If have hint: | |
498 | * check if hint fits in good => OK | |
499 | * check if hint fits in good | free => convert free | |
500 | * Otherwise: | |
501 | * search in good, found => OK | |
502 | * search in good | free, found => convert free | |
503 | * search in good | compat | free, found => convert free. | |
504 | */ | |
d0f13e3c | 505 | |
3a8247cc PM |
506 | #ifdef CONFIG_PPC_64K_PAGES |
507 | /* If we support combo pages, we can allow 64k pages in 4k slices */ | |
508 | if (psize == MMU_PAGE_64K) { | |
7a06c668 | 509 | slice_mask_for_size(mm, MMU_PAGE_4K, &compat_mask, high_limit); |
3a8247cc | 510 | if (fixed) |
f3207c12 | 511 | slice_or_mask(&good_mask, &compat_mask); |
3a8247cc PM |
512 | } |
513 | #endif | |
d0f13e3c | 514 | |
3a8247cc PM |
515 | /* First check hint if it's valid or if we have MAP_FIXED */ |
516 | if (addr != 0 || fixed) { | |
d0f13e3c | 517 | /* Build a mask for the requested range */ |
a4d36215 | 518 | slice_range_to_mask(addr, len, &mask); |
d0f13e3c BH |
519 | slice_print_mask(" mask", mask); |
520 | ||
521 | /* Check if we fit in the good mask. If we do, we just return, | |
522 | * nothing else to do | |
523 | */ | |
957b778a | 524 | if (slice_check_fit(mm, mask, good_mask)) { |
d0f13e3c BH |
525 | slice_dbg(" fits good !\n"); |
526 | return addr; | |
527 | } | |
3a8247cc PM |
528 | } else { |
529 | /* Now let's see if we can find something in the existing | |
530 | * slices for that size | |
d0f13e3c | 531 | */ |
f4ea6dcb AK |
532 | newaddr = slice_find_area(mm, len, good_mask, |
533 | psize, topdown, high_limit); | |
3a8247cc PM |
534 | if (newaddr != -ENOMEM) { |
535 | /* Found within the good mask, we don't have to setup, | |
536 | * we thus return directly | |
537 | */ | |
538 | slice_dbg(" found area at 0x%lx\n", newaddr); | |
539 | return newaddr; | |
d0f13e3c BH |
540 | } |
541 | } | |
7a06c668 AK |
542 | /* |
543 | * We don't fit in the good mask, check what other slices are | |
3a8247cc PM |
544 | * empty and thus can be converted |
545 | */ | |
7a06c668 | 546 | slice_mask_for_free(mm, &potential_mask, high_limit); |
f3207c12 | 547 | slice_or_mask(&potential_mask, &good_mask); |
3a8247cc PM |
548 | slice_print_mask(" potential", potential_mask); |
549 | ||
957b778a | 550 | if ((addr != 0 || fixed) && slice_check_fit(mm, mask, potential_mask)) { |
3a8247cc PM |
551 | slice_dbg(" fits potential !\n"); |
552 | goto convert; | |
553 | } | |
554 | ||
555 | /* If we have MAP_FIXED and failed the above steps, then error out */ | |
d0f13e3c BH |
556 | if (fixed) |
557 | return -EBUSY; | |
558 | ||
d0f13e3c BH |
559 | slice_dbg(" search...\n"); |
560 | ||
3a8247cc PM |
561 | /* If we had a hint that didn't work out, see if we can fit |
562 | * anywhere in the good area. | |
d0f13e3c | 563 | */ |
3a8247cc | 564 | if (addr) { |
f4ea6dcb AK |
565 | addr = slice_find_area(mm, len, good_mask, |
566 | psize, topdown, high_limit); | |
3a8247cc PM |
567 | if (addr != -ENOMEM) { |
568 | slice_dbg(" found area at 0x%lx\n", addr); | |
569 | return addr; | |
570 | } | |
d0f13e3c BH |
571 | } |
572 | ||
573 | /* Now let's see if we can find something in the existing slices | |
3a8247cc | 574 | * for that size plus free slices |
d0f13e3c | 575 | */ |
f4ea6dcb AK |
576 | addr = slice_find_area(mm, len, potential_mask, |
577 | psize, topdown, high_limit); | |
3a8247cc PM |
578 | |
579 | #ifdef CONFIG_PPC_64K_PAGES | |
580 | if (addr == -ENOMEM && psize == MMU_PAGE_64K) { | |
581 | /* retry the search with 4k-page slices included */ | |
f3207c12 | 582 | slice_or_mask(&potential_mask, &compat_mask); |
f4ea6dcb AK |
583 | addr = slice_find_area(mm, len, potential_mask, |
584 | psize, topdown, high_limit); | |
3a8247cc PM |
585 | } |
586 | #endif | |
587 | ||
d0f13e3c BH |
588 | if (addr == -ENOMEM) |
589 | return -ENOMEM; | |
590 | ||
a4d36215 | 591 | slice_range_to_mask(addr, len, &mask); |
d0f13e3c BH |
592 | slice_dbg(" found potential area at 0x%lx\n", addr); |
593 | slice_print_mask(" mask", mask); | |
594 | ||
595 | convert: | |
f3207c12 AK |
596 | slice_andnot_mask(&mask, &good_mask); |
597 | slice_andnot_mask(&mask, &compat_mask); | |
598 | if (mask.low_slices || !bitmap_empty(mask.high_slices, SLICE_NUM_HIGH)) { | |
3a8247cc PM |
599 | slice_convert(mm, mask, psize); |
600 | if (psize > MMU_PAGE_BASE) | |
84c3d4aa | 601 | on_each_cpu(slice_flush_segments, mm, 1); |
3a8247cc | 602 | } |
d0f13e3c BH |
603 | return addr; |
604 | ||
605 | } | |
606 | EXPORT_SYMBOL_GPL(slice_get_unmapped_area); | |
607 | ||
608 | unsigned long arch_get_unmapped_area(struct file *filp, | |
609 | unsigned long addr, | |
610 | unsigned long len, | |
611 | unsigned long pgoff, | |
612 | unsigned long flags) | |
613 | { | |
614 | return slice_get_unmapped_area(addr, len, flags, | |
34d07177 | 615 | current->mm->context.user_psize, 0); |
d0f13e3c BH |
616 | } |
617 | ||
618 | unsigned long arch_get_unmapped_area_topdown(struct file *filp, | |
619 | const unsigned long addr0, | |
620 | const unsigned long len, | |
621 | const unsigned long pgoff, | |
622 | const unsigned long flags) | |
623 | { | |
624 | return slice_get_unmapped_area(addr0, len, flags, | |
34d07177 | 625 | current->mm->context.user_psize, 1); |
d0f13e3c BH |
626 | } |
627 | ||
628 | unsigned int get_slice_psize(struct mm_struct *mm, unsigned long addr) | |
629 | { | |
7aa0727f AK |
630 | unsigned char *hpsizes; |
631 | int index, mask_index; | |
d0f13e3c | 632 | |
764041e0 AK |
633 | /* |
634 | * Radix doesn't use slice, but can get enabled along with MMU_SLICE | |
635 | */ | |
636 | if (radix_enabled()) { | |
637 | #ifdef CONFIG_PPC_64K_PAGES | |
638 | return MMU_PAGE_64K; | |
639 | #else | |
640 | return MMU_PAGE_4K; | |
641 | #endif | |
642 | } | |
d0f13e3c | 643 | if (addr < SLICE_LOW_TOP) { |
7aa0727f AK |
644 | u64 lpsizes; |
645 | lpsizes = mm->context.low_slices_psize; | |
d0f13e3c | 646 | index = GET_LOW_SLICE_INDEX(addr); |
7aa0727f | 647 | return (lpsizes >> (index * 4)) & 0xf; |
d0f13e3c | 648 | } |
7aa0727f AK |
649 | hpsizes = mm->context.high_slices_psize; |
650 | index = GET_HIGH_SLICE_INDEX(addr); | |
651 | mask_index = index & 0x1; | |
652 | return (hpsizes[index >> 1] >> (mask_index * 4)) & 0xf; | |
d0f13e3c BH |
653 | } |
654 | EXPORT_SYMBOL_GPL(get_slice_psize); | |
655 | ||
656 | /* | |
657 | * This is called by hash_page when it needs to do a lazy conversion of | |
658 | * an address space from real 64K pages to combo 4K pages (typically | |
659 | * when hitting a non cacheable mapping on a processor or hypervisor | |
660 | * that won't allow them for 64K pages). | |
661 | * | |
662 | * This is also called in init_new_context() to change back the user | |
663 | * psize from whatever the parent context had it set to | |
9dfe5c53 | 664 | * N.B. This may be called before mm->context.id has been set. |
d0f13e3c BH |
665 | * |
666 | * This function will only change the content of the {low,high)_slice_psize | |
667 | * masks, it will not flush SLBs as this shall be handled lazily by the | |
668 | * caller. | |
669 | */ | |
670 | void slice_set_user_psize(struct mm_struct *mm, unsigned int psize) | |
671 | { | |
7aa0727f AK |
672 | int index, mask_index; |
673 | unsigned char *hpsizes; | |
674 | unsigned long flags, lpsizes; | |
d0f13e3c BH |
675 | unsigned int old_psize; |
676 | int i; | |
677 | ||
678 | slice_dbg("slice_set_user_psize(mm=%p, psize=%d)\n", mm, psize); | |
679 | ||
764041e0 | 680 | VM_BUG_ON(radix_enabled()); |
d0f13e3c BH |
681 | spin_lock_irqsave(&slice_convert_lock, flags); |
682 | ||
683 | old_psize = mm->context.user_psize; | |
684 | slice_dbg(" old_psize=%d\n", old_psize); | |
685 | if (old_psize == psize) | |
686 | goto bail; | |
687 | ||
688 | mm->context.user_psize = psize; | |
689 | wmb(); | |
690 | ||
691 | lpsizes = mm->context.low_slices_psize; | |
692 | for (i = 0; i < SLICE_NUM_LOW; i++) | |
693 | if (((lpsizes >> (i * 4)) & 0xf) == old_psize) | |
694 | lpsizes = (lpsizes & ~(0xful << (i * 4))) | | |
695 | (((unsigned long)psize) << (i * 4)); | |
7aa0727f AK |
696 | /* Assign the value back */ |
697 | mm->context.low_slices_psize = lpsizes; | |
d0f13e3c BH |
698 | |
699 | hpsizes = mm->context.high_slices_psize; | |
7aa0727f AK |
700 | for (i = 0; i < SLICE_NUM_HIGH; i++) { |
701 | mask_index = i & 0x1; | |
702 | index = i >> 1; | |
703 | if (((hpsizes[index] >> (mask_index * 4)) & 0xf) == old_psize) | |
704 | hpsizes[index] = (hpsizes[index] & | |
705 | ~(0xf << (mask_index * 4))) | | |
706 | (((unsigned long)psize) << (mask_index * 4)); | |
707 | } | |
708 | ||
709 | ||
d0f13e3c | 710 | |
d0f13e3c BH |
711 | |
712 | slice_dbg(" lsps=%lx, hsps=%lx\n", | |
302413ca AK |
713 | (unsigned long)mm->context.low_slices_psize, |
714 | (unsigned long)mm->context.high_slices_psize); | |
d0f13e3c BH |
715 | |
716 | bail: | |
717 | spin_unlock_irqrestore(&slice_convert_lock, flags); | |
718 | } | |
719 | ||
3a8247cc PM |
720 | void slice_set_range_psize(struct mm_struct *mm, unsigned long start, |
721 | unsigned long len, unsigned int psize) | |
722 | { | |
a4d36215 | 723 | struct slice_mask mask; |
3a8247cc | 724 | |
764041e0 | 725 | VM_BUG_ON(radix_enabled()); |
a4d36215 AK |
726 | |
727 | slice_range_to_mask(start, len, &mask); | |
3a8247cc PM |
728 | slice_convert(mm, mask, psize); |
729 | } | |
730 | ||
6643773c | 731 | #ifdef CONFIG_HUGETLB_PAGE |
d0f13e3c | 732 | /* |
48fc7f7e | 733 | * is_hugepage_only_range() is used by generic code to verify whether |
d0f13e3c BH |
734 | * a normal mmap mapping (non hugetlbfs) is valid on a given area. |
735 | * | |
736 | * until the generic code provides a more generic hook and/or starts | |
737 | * calling arch get_unmapped_area for MAP_FIXED (which our implementation | |
738 | * here knows how to deal with), we hijack it to keep standard mappings | |
739 | * away from us. | |
740 | * | |
741 | * because of that generic code limitation, MAP_FIXED mapping cannot | |
742 | * "convert" back a slice with no VMAs to the standard page size, only | |
743 | * get_unmapped_area() can. It would be possible to fix it here but I | |
744 | * prefer working on fixing the generic code instead. | |
745 | * | |
746 | * WARNING: This will not work if hugetlbfs isn't enabled since the | |
747 | * generic code will redefine that function as 0 in that. This is ok | |
748 | * for now as we only use slices with hugetlbfs enabled. This should | |
749 | * be fixed as the generic code gets fixed. | |
750 | */ | |
751 | int is_hugepage_only_range(struct mm_struct *mm, unsigned long addr, | |
752 | unsigned long len) | |
753 | { | |
754 | struct slice_mask mask, available; | |
9ba0fdbf | 755 | unsigned int psize = mm->context.user_psize; |
7a06c668 | 756 | unsigned long high_limit = mm->context.slb_addr_limit; |
d0f13e3c | 757 | |
764041e0 AK |
758 | if (radix_enabled()) |
759 | return 0; | |
760 | ||
a4d36215 | 761 | slice_range_to_mask(addr, len, &mask); |
7a06c668 | 762 | slice_mask_for_size(mm, psize, &available, high_limit); |
9ba0fdbf DK |
763 | #ifdef CONFIG_PPC_64K_PAGES |
764 | /* We need to account for 4k slices too */ | |
765 | if (psize == MMU_PAGE_64K) { | |
766 | struct slice_mask compat_mask; | |
7a06c668 | 767 | slice_mask_for_size(mm, MMU_PAGE_4K, &compat_mask, high_limit); |
f3207c12 | 768 | slice_or_mask(&available, &compat_mask); |
9ba0fdbf DK |
769 | } |
770 | #endif | |
d0f13e3c BH |
771 | |
772 | #if 0 /* too verbose */ | |
773 | slice_dbg("is_hugepage_only_range(mm=%p, addr=%lx, len=%lx)\n", | |
774 | mm, addr, len); | |
775 | slice_print_mask(" mask", mask); | |
776 | slice_print_mask(" available", available); | |
777 | #endif | |
957b778a | 778 | return !slice_check_fit(mm, mask, available); |
d0f13e3c | 779 | } |
6643773c | 780 | #endif |