Loading...
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * address space "slices" (meta-segments) support
4 *
5 * Copyright (C) 2007 Benjamin Herrenschmidt, IBM Corporation.
6 *
7 * Based on hugetlb implementation
8 *
9 * Copyright (C) 2003 David Gibson, IBM Corporation.
10 */
11
12#undef DEBUG
13
14#include <linux/kernel.h>
15#include <linux/mm.h>
16#include <linux/pagemap.h>
17#include <linux/err.h>
18#include <linux/spinlock.h>
19#include <linux/export.h>
20#include <linux/hugetlb.h>
21#include <linux/sched/mm.h>
22#include <linux/security.h>
23#include <asm/mman.h>
24#include <asm/mmu.h>
25#include <asm/copro.h>
26#include <asm/hugetlb.h>
27#include <asm/mmu_context.h>
28
29static DEFINE_SPINLOCK(slice_convert_lock);
30
31#ifdef DEBUG
32int _slice_debug = 1;
33
34static void slice_print_mask(const char *label, const struct slice_mask *mask)
35{
36 if (!_slice_debug)
37 return;
38 pr_devel("%s low_slice: %*pbl\n", label,
39 (int)SLICE_NUM_LOW, &mask->low_slices);
40 pr_devel("%s high_slice: %*pbl\n", label,
41 (int)SLICE_NUM_HIGH, mask->high_slices);
42}
43
44#define slice_dbg(fmt...) do { if (_slice_debug) pr_devel(fmt); } while (0)
45
46#else
47
48static void slice_print_mask(const char *label, const struct slice_mask *mask) {}
49#define slice_dbg(fmt...)
50
51#endif
52
53static inline bool slice_addr_is_low(unsigned long addr)
54{
55 u64 tmp = (u64)addr;
56
57 return tmp < SLICE_LOW_TOP;
58}
59
60static void slice_range_to_mask(unsigned long start, unsigned long len,
61 struct slice_mask *ret)
62{
63 unsigned long end = start + len - 1;
64
65 ret->low_slices = 0;
66 if (SLICE_NUM_HIGH)
67 bitmap_zero(ret->high_slices, SLICE_NUM_HIGH);
68
69 if (slice_addr_is_low(start)) {
70 unsigned long mend = min(end,
71 (unsigned long)(SLICE_LOW_TOP - 1));
72
73 ret->low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1))
74 - (1u << GET_LOW_SLICE_INDEX(start));
75 }
76
77 if (SLICE_NUM_HIGH && !slice_addr_is_low(end)) {
78 unsigned long start_index = GET_HIGH_SLICE_INDEX(start);
79 unsigned long align_end = ALIGN(end, (1UL << SLICE_HIGH_SHIFT));
80 unsigned long count = GET_HIGH_SLICE_INDEX(align_end) - start_index;
81
82 bitmap_set(ret->high_slices, start_index, count);
83 }
84}
85
86static int slice_area_is_free(struct mm_struct *mm, unsigned long addr,
87 unsigned long len)
88{
89 struct vm_area_struct *vma;
90
91 if ((mm_ctx_slb_addr_limit(&mm->context) - len) < addr)
92 return 0;
93 vma = find_vma(mm, addr);
94 return (!vma || (addr + len) <= vm_start_gap(vma));
95}
96
97static int slice_low_has_vma(struct mm_struct *mm, unsigned long slice)
98{
99 return !slice_area_is_free(mm, slice << SLICE_LOW_SHIFT,
100 1ul << SLICE_LOW_SHIFT);
101}
102
103static int slice_high_has_vma(struct mm_struct *mm, unsigned long slice)
104{
105 unsigned long start = slice << SLICE_HIGH_SHIFT;
106 unsigned long end = start + (1ul << SLICE_HIGH_SHIFT);
107
108 /* Hack, so that each addresses is controlled by exactly one
109 * of the high or low area bitmaps, the first high area starts
110 * at 4GB, not 0 */
111 if (start == 0)
112 start = (unsigned long)SLICE_LOW_TOP;
113
114 return !slice_area_is_free(mm, start, end - start);
115}
116
117static void slice_mask_for_free(struct mm_struct *mm, struct slice_mask *ret,
118 unsigned long high_limit)
119{
120 unsigned long i;
121
122 ret->low_slices = 0;
123 if (SLICE_NUM_HIGH)
124 bitmap_zero(ret->high_slices, SLICE_NUM_HIGH);
125
126 for (i = 0; i < SLICE_NUM_LOW; i++)
127 if (!slice_low_has_vma(mm, i))
128 ret->low_slices |= 1u << i;
129
130 if (slice_addr_is_low(high_limit - 1))
131 return;
132
133 for (i = 0; i < GET_HIGH_SLICE_INDEX(high_limit); i++)
134 if (!slice_high_has_vma(mm, i))
135 __set_bit(i, ret->high_slices);
136}
137
138static bool slice_check_range_fits(struct mm_struct *mm,
139 const struct slice_mask *available,
140 unsigned long start, unsigned long len)
141{
142 unsigned long end = start + len - 1;
143 u64 low_slices = 0;
144
145 if (slice_addr_is_low(start)) {
146 unsigned long mend = min(end,
147 (unsigned long)(SLICE_LOW_TOP - 1));
148
149 low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1))
150 - (1u << GET_LOW_SLICE_INDEX(start));
151 }
152 if ((low_slices & available->low_slices) != low_slices)
153 return false;
154
155 if (SLICE_NUM_HIGH && !slice_addr_is_low(end)) {
156 unsigned long start_index = GET_HIGH_SLICE_INDEX(start);
157 unsigned long align_end = ALIGN(end, (1UL << SLICE_HIGH_SHIFT));
158 unsigned long count = GET_HIGH_SLICE_INDEX(align_end) - start_index;
159 unsigned long i;
160
161 for (i = start_index; i < start_index + count; i++) {
162 if (!test_bit(i, available->high_slices))
163 return false;
164 }
165 }
166
167 return true;
168}
169
170static void slice_flush_segments(void *parm)
171{
172#ifdef CONFIG_PPC64
173 struct mm_struct *mm = parm;
174 unsigned long flags;
175
176 if (mm != current->active_mm)
177 return;
178
179 copy_mm_to_paca(current->active_mm);
180
181 local_irq_save(flags);
182 slb_flush_and_restore_bolted();
183 local_irq_restore(flags);
184#endif
185}
186
187static void slice_convert(struct mm_struct *mm,
188 const struct slice_mask *mask, int psize)
189{
190 int index, mask_index;
191 /* Write the new slice psize bits */
192 unsigned char *hpsizes, *lpsizes;
193 struct slice_mask *psize_mask, *old_mask;
194 unsigned long i, flags;
195 int old_psize;
196
197 slice_dbg("slice_convert(mm=%p, psize=%d)\n", mm, psize);
198 slice_print_mask(" mask", mask);
199
200 psize_mask = slice_mask_for_size(&mm->context, psize);
201
202 /* We need to use a spinlock here to protect against
203 * concurrent 64k -> 4k demotion ...
204 */
205 spin_lock_irqsave(&slice_convert_lock, flags);
206
207 lpsizes = mm_ctx_low_slices(&mm->context);
208 for (i = 0; i < SLICE_NUM_LOW; i++) {
209 if (!(mask->low_slices & (1u << i)))
210 continue;
211
212 mask_index = i & 0x1;
213 index = i >> 1;
214
215 /* Update the slice_mask */
216 old_psize = (lpsizes[index] >> (mask_index * 4)) & 0xf;
217 old_mask = slice_mask_for_size(&mm->context, old_psize);
218 old_mask->low_slices &= ~(1u << i);
219 psize_mask->low_slices |= 1u << i;
220
221 /* Update the sizes array */
222 lpsizes[index] = (lpsizes[index] & ~(0xf << (mask_index * 4))) |
223 (((unsigned long)psize) << (mask_index * 4));
224 }
225
226 hpsizes = mm_ctx_high_slices(&mm->context);
227 for (i = 0; i < GET_HIGH_SLICE_INDEX(mm_ctx_slb_addr_limit(&mm->context)); i++) {
228 if (!test_bit(i, mask->high_slices))
229 continue;
230
231 mask_index = i & 0x1;
232 index = i >> 1;
233
234 /* Update the slice_mask */
235 old_psize = (hpsizes[index] >> (mask_index * 4)) & 0xf;
236 old_mask = slice_mask_for_size(&mm->context, old_psize);
237 __clear_bit(i, old_mask->high_slices);
238 __set_bit(i, psize_mask->high_slices);
239
240 /* Update the sizes array */
241 hpsizes[index] = (hpsizes[index] & ~(0xf << (mask_index * 4))) |
242 (((unsigned long)psize) << (mask_index * 4));
243 }
244
245 slice_dbg(" lsps=%lx, hsps=%lx\n",
246 (unsigned long)mm_ctx_low_slices(&mm->context),
247 (unsigned long)mm_ctx_high_slices(&mm->context));
248
249 spin_unlock_irqrestore(&slice_convert_lock, flags);
250
251 copro_flush_all_slbs(mm);
252}
253
254/*
255 * Compute which slice addr is part of;
256 * set *boundary_addr to the start or end boundary of that slice
257 * (depending on 'end' parameter);
258 * return boolean indicating if the slice is marked as available in the
259 * 'available' slice_mark.
260 */
261static bool slice_scan_available(unsigned long addr,
262 const struct slice_mask *available,
263 int end, unsigned long *boundary_addr)
264{
265 unsigned long slice;
266 if (slice_addr_is_low(addr)) {
267 slice = GET_LOW_SLICE_INDEX(addr);
268 *boundary_addr = (slice + end) << SLICE_LOW_SHIFT;
269 return !!(available->low_slices & (1u << slice));
270 } else {
271 slice = GET_HIGH_SLICE_INDEX(addr);
272 *boundary_addr = (slice + end) ?
273 ((slice + end) << SLICE_HIGH_SHIFT) : SLICE_LOW_TOP;
274 return !!test_bit(slice, available->high_slices);
275 }
276}
277
278static unsigned long slice_find_area_bottomup(struct mm_struct *mm,
279 unsigned long len,
280 const struct slice_mask *available,
281 int psize, unsigned long high_limit)
282{
283 int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
284 unsigned long addr, found, next_end;
285 struct vm_unmapped_area_info info;
286
287 info.flags = 0;
288 info.length = len;
289 info.align_mask = PAGE_MASK & ((1ul << pshift) - 1);
290 info.align_offset = 0;
291
292 addr = TASK_UNMAPPED_BASE;
293 /*
294 * Check till the allow max value for this mmap request
295 */
296 while (addr < high_limit) {
297 info.low_limit = addr;
298 if (!slice_scan_available(addr, available, 1, &addr))
299 continue;
300
301 next_slice:
302 /*
303 * At this point [info.low_limit; addr) covers
304 * available slices only and ends at a slice boundary.
305 * Check if we need to reduce the range, or if we can
306 * extend it to cover the next available slice.
307 */
308 if (addr >= high_limit)
309 addr = high_limit;
310 else if (slice_scan_available(addr, available, 1, &next_end)) {
311 addr = next_end;
312 goto next_slice;
313 }
314 info.high_limit = addr;
315
316 found = vm_unmapped_area(&info);
317 if (!(found & ~PAGE_MASK))
318 return found;
319 }
320
321 return -ENOMEM;
322}
323
324static unsigned long slice_find_area_topdown(struct mm_struct *mm,
325 unsigned long len,
326 const struct slice_mask *available,
327 int psize, unsigned long high_limit)
328{
329 int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
330 unsigned long addr, found, prev;
331 struct vm_unmapped_area_info info;
332 unsigned long min_addr = max(PAGE_SIZE, mmap_min_addr);
333
334 info.flags = VM_UNMAPPED_AREA_TOPDOWN;
335 info.length = len;
336 info.align_mask = PAGE_MASK & ((1ul << pshift) - 1);
337 info.align_offset = 0;
338
339 addr = mm->mmap_base;
340 /*
341 * If we are trying to allocate above DEFAULT_MAP_WINDOW
342 * Add the different to the mmap_base.
343 * Only for that request for which high_limit is above
344 * DEFAULT_MAP_WINDOW we should apply this.
345 */
346 if (high_limit > DEFAULT_MAP_WINDOW)
347 addr += mm_ctx_slb_addr_limit(&mm->context) - DEFAULT_MAP_WINDOW;
348
349 while (addr > min_addr) {
350 info.high_limit = addr;
351 if (!slice_scan_available(addr - 1, available, 0, &addr))
352 continue;
353
354 prev_slice:
355 /*
356 * At this point [addr; info.high_limit) covers
357 * available slices only and starts at a slice boundary.
358 * Check if we need to reduce the range, or if we can
359 * extend it to cover the previous available slice.
360 */
361 if (addr < min_addr)
362 addr = min_addr;
363 else if (slice_scan_available(addr - 1, available, 0, &prev)) {
364 addr = prev;
365 goto prev_slice;
366 }
367 info.low_limit = addr;
368
369 found = vm_unmapped_area(&info);
370 if (!(found & ~PAGE_MASK))
371 return found;
372 }
373
374 /*
375 * A failed mmap() very likely causes application failure,
376 * so fall back to the bottom-up function here. This scenario
377 * can happen with large stack limits and large mmap()
378 * allocations.
379 */
380 return slice_find_area_bottomup(mm, len, available, psize, high_limit);
381}
382
383
384static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len,
385 const struct slice_mask *mask, int psize,
386 int topdown, unsigned long high_limit)
387{
388 if (topdown)
389 return slice_find_area_topdown(mm, len, mask, psize, high_limit);
390 else
391 return slice_find_area_bottomup(mm, len, mask, psize, high_limit);
392}
393
394static inline void slice_copy_mask(struct slice_mask *dst,
395 const struct slice_mask *src)
396{
397 dst->low_slices = src->low_slices;
398 if (!SLICE_NUM_HIGH)
399 return;
400 bitmap_copy(dst->high_slices, src->high_slices, SLICE_NUM_HIGH);
401}
402
403static inline void slice_or_mask(struct slice_mask *dst,
404 const struct slice_mask *src1,
405 const struct slice_mask *src2)
406{
407 dst->low_slices = src1->low_slices | src2->low_slices;
408 if (!SLICE_NUM_HIGH)
409 return;
410 bitmap_or(dst->high_slices, src1->high_slices, src2->high_slices, SLICE_NUM_HIGH);
411}
412
413static inline void slice_andnot_mask(struct slice_mask *dst,
414 const struct slice_mask *src1,
415 const struct slice_mask *src2)
416{
417 dst->low_slices = src1->low_slices & ~src2->low_slices;
418 if (!SLICE_NUM_HIGH)
419 return;
420 bitmap_andnot(dst->high_slices, src1->high_slices, src2->high_slices, SLICE_NUM_HIGH);
421}
422
423#ifdef CONFIG_PPC_64K_PAGES
424#define MMU_PAGE_BASE MMU_PAGE_64K
425#else
426#define MMU_PAGE_BASE MMU_PAGE_4K
427#endif
428
429unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
430 unsigned long flags, unsigned int psize,
431 int topdown)
432{
433 struct slice_mask good_mask;
434 struct slice_mask potential_mask;
435 const struct slice_mask *maskp;
436 const struct slice_mask *compat_maskp = NULL;
437 int fixed = (flags & MAP_FIXED);
438 int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
439 unsigned long page_size = 1UL << pshift;
440 struct mm_struct *mm = current->mm;
441 unsigned long newaddr;
442 unsigned long high_limit;
443
444 high_limit = DEFAULT_MAP_WINDOW;
445 if (addr >= high_limit || (fixed && (addr + len > high_limit)))
446 high_limit = TASK_SIZE;
447
448 if (len > high_limit)
449 return -ENOMEM;
450 if (len & (page_size - 1))
451 return -EINVAL;
452 if (fixed) {
453 if (addr & (page_size - 1))
454 return -EINVAL;
455 if (addr > high_limit - len)
456 return -ENOMEM;
457 }
458
459 if (high_limit > mm_ctx_slb_addr_limit(&mm->context)) {
460 /*
461 * Increasing the slb_addr_limit does not require
462 * slice mask cache to be recalculated because it should
463 * be already initialised beyond the old address limit.
464 */
465 mm_ctx_set_slb_addr_limit(&mm->context, high_limit);
466
467 on_each_cpu(slice_flush_segments, mm, 1);
468 }
469
470 /* Sanity checks */
471 BUG_ON(mm->task_size == 0);
472 BUG_ON(mm_ctx_slb_addr_limit(&mm->context) == 0);
473 VM_BUG_ON(radix_enabled());
474
475 slice_dbg("slice_get_unmapped_area(mm=%p, psize=%d...\n", mm, psize);
476 slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d\n",
477 addr, len, flags, topdown);
478
479 /* If hint, make sure it matches our alignment restrictions */
480 if (!fixed && addr) {
481 addr = _ALIGN_UP(addr, page_size);
482 slice_dbg(" aligned addr=%lx\n", addr);
483 /* Ignore hint if it's too large or overlaps a VMA */
484 if (addr > high_limit - len || addr < mmap_min_addr ||
485 !slice_area_is_free(mm, addr, len))
486 addr = 0;
487 }
488
489 /* First make up a "good" mask of slices that have the right size
490 * already
491 */
492 maskp = slice_mask_for_size(&mm->context, psize);
493
494 /*
495 * Here "good" means slices that are already the right page size,
496 * "compat" means slices that have a compatible page size (i.e.
497 * 4k in a 64k pagesize kernel), and "free" means slices without
498 * any VMAs.
499 *
500 * If MAP_FIXED:
501 * check if fits in good | compat => OK
502 * check if fits in good | compat | free => convert free
503 * else bad
504 * If have hint:
505 * check if hint fits in good => OK
506 * check if hint fits in good | free => convert free
507 * Otherwise:
508 * search in good, found => OK
509 * search in good | free, found => convert free
510 * search in good | compat | free, found => convert free.
511 */
512
513 /*
514 * If we support combo pages, we can allow 64k pages in 4k slices
515 * The mask copies could be avoided in most cases here if we had
516 * a pointer to good mask for the next code to use.
517 */
518 if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && psize == MMU_PAGE_64K) {
519 compat_maskp = slice_mask_for_size(&mm->context, MMU_PAGE_4K);
520 if (fixed)
521 slice_or_mask(&good_mask, maskp, compat_maskp);
522 else
523 slice_copy_mask(&good_mask, maskp);
524 } else {
525 slice_copy_mask(&good_mask, maskp);
526 }
527
528 slice_print_mask(" good_mask", &good_mask);
529 if (compat_maskp)
530 slice_print_mask(" compat_mask", compat_maskp);
531
532 /* First check hint if it's valid or if we have MAP_FIXED */
533 if (addr != 0 || fixed) {
534 /* Check if we fit in the good mask. If we do, we just return,
535 * nothing else to do
536 */
537 if (slice_check_range_fits(mm, &good_mask, addr, len)) {
538 slice_dbg(" fits good !\n");
539 newaddr = addr;
540 goto return_addr;
541 }
542 } else {
543 /* Now let's see if we can find something in the existing
544 * slices for that size
545 */
546 newaddr = slice_find_area(mm, len, &good_mask,
547 psize, topdown, high_limit);
548 if (newaddr != -ENOMEM) {
549 /* Found within the good mask, we don't have to setup,
550 * we thus return directly
551 */
552 slice_dbg(" found area at 0x%lx\n", newaddr);
553 goto return_addr;
554 }
555 }
556 /*
557 * We don't fit in the good mask, check what other slices are
558 * empty and thus can be converted
559 */
560 slice_mask_for_free(mm, &potential_mask, high_limit);
561 slice_or_mask(&potential_mask, &potential_mask, &good_mask);
562 slice_print_mask(" potential", &potential_mask);
563
564 if (addr != 0 || fixed) {
565 if (slice_check_range_fits(mm, &potential_mask, addr, len)) {
566 slice_dbg(" fits potential !\n");
567 newaddr = addr;
568 goto convert;
569 }
570 }
571
572 /* If we have MAP_FIXED and failed the above steps, then error out */
573 if (fixed)
574 return -EBUSY;
575
576 slice_dbg(" search...\n");
577
578 /* If we had a hint that didn't work out, see if we can fit
579 * anywhere in the good area.
580 */
581 if (addr) {
582 newaddr = slice_find_area(mm, len, &good_mask,
583 psize, topdown, high_limit);
584 if (newaddr != -ENOMEM) {
585 slice_dbg(" found area at 0x%lx\n", newaddr);
586 goto return_addr;
587 }
588 }
589
590 /* Now let's see if we can find something in the existing slices
591 * for that size plus free slices
592 */
593 newaddr = slice_find_area(mm, len, &potential_mask,
594 psize, topdown, high_limit);
595
596 if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && newaddr == -ENOMEM &&
597 psize == MMU_PAGE_64K) {
598 /* retry the search with 4k-page slices included */
599 slice_or_mask(&potential_mask, &potential_mask, compat_maskp);
600 newaddr = slice_find_area(mm, len, &potential_mask,
601 psize, topdown, high_limit);
602 }
603
604 if (newaddr == -ENOMEM)
605 return -ENOMEM;
606
607 slice_range_to_mask(newaddr, len, &potential_mask);
608 slice_dbg(" found potential area at 0x%lx\n", newaddr);
609 slice_print_mask(" mask", &potential_mask);
610
611 convert:
612 /*
613 * Try to allocate the context before we do slice convert
614 * so that we handle the context allocation failure gracefully.
615 */
616 if (need_extra_context(mm, newaddr)) {
617 if (alloc_extended_context(mm, newaddr) < 0)
618 return -ENOMEM;
619 }
620
621 slice_andnot_mask(&potential_mask, &potential_mask, &good_mask);
622 if (compat_maskp && !fixed)
623 slice_andnot_mask(&potential_mask, &potential_mask, compat_maskp);
624 if (potential_mask.low_slices ||
625 (SLICE_NUM_HIGH &&
626 !bitmap_empty(potential_mask.high_slices, SLICE_NUM_HIGH))) {
627 slice_convert(mm, &potential_mask, psize);
628 if (psize > MMU_PAGE_BASE)
629 on_each_cpu(slice_flush_segments, mm, 1);
630 }
631 return newaddr;
632
633return_addr:
634 if (need_extra_context(mm, newaddr)) {
635 if (alloc_extended_context(mm, newaddr) < 0)
636 return -ENOMEM;
637 }
638 return newaddr;
639}
640EXPORT_SYMBOL_GPL(slice_get_unmapped_area);
641
642unsigned long arch_get_unmapped_area(struct file *filp,
643 unsigned long addr,
644 unsigned long len,
645 unsigned long pgoff,
646 unsigned long flags)
647{
648 return slice_get_unmapped_area(addr, len, flags,
649 mm_ctx_user_psize(¤t->mm->context), 0);
650}
651
652unsigned long arch_get_unmapped_area_topdown(struct file *filp,
653 const unsigned long addr0,
654 const unsigned long len,
655 const unsigned long pgoff,
656 const unsigned long flags)
657{
658 return slice_get_unmapped_area(addr0, len, flags,
659 mm_ctx_user_psize(¤t->mm->context), 1);
660}
661
662unsigned int get_slice_psize(struct mm_struct *mm, unsigned long addr)
663{
664 unsigned char *psizes;
665 int index, mask_index;
666
667 VM_BUG_ON(radix_enabled());
668
669 if (slice_addr_is_low(addr)) {
670 psizes = mm_ctx_low_slices(&mm->context);
671 index = GET_LOW_SLICE_INDEX(addr);
672 } else {
673 psizes = mm_ctx_high_slices(&mm->context);
674 index = GET_HIGH_SLICE_INDEX(addr);
675 }
676 mask_index = index & 0x1;
677 return (psizes[index >> 1] >> (mask_index * 4)) & 0xf;
678}
679EXPORT_SYMBOL_GPL(get_slice_psize);
680
681void slice_init_new_context_exec(struct mm_struct *mm)
682{
683 unsigned char *hpsizes, *lpsizes;
684 struct slice_mask *mask;
685 unsigned int psize = mmu_virtual_psize;
686
687 slice_dbg("slice_init_new_context_exec(mm=%p)\n", mm);
688
689 /*
690 * In the case of exec, use the default limit. In the
691 * case of fork it is just inherited from the mm being
692 * duplicated.
693 */
694 mm_ctx_set_slb_addr_limit(&mm->context, SLB_ADDR_LIMIT_DEFAULT);
695 mm_ctx_set_user_psize(&mm->context, psize);
696
697 /*
698 * Set all slice psizes to the default.
699 */
700 lpsizes = mm_ctx_low_slices(&mm->context);
701 memset(lpsizes, (psize << 4) | psize, SLICE_NUM_LOW >> 1);
702
703 hpsizes = mm_ctx_high_slices(&mm->context);
704 memset(hpsizes, (psize << 4) | psize, SLICE_NUM_HIGH >> 1);
705
706 /*
707 * Slice mask cache starts zeroed, fill the default size cache.
708 */
709 mask = slice_mask_for_size(&mm->context, psize);
710 mask->low_slices = ~0UL;
711 if (SLICE_NUM_HIGH)
712 bitmap_fill(mask->high_slices, SLICE_NUM_HIGH);
713}
714
715#ifdef CONFIG_PPC_BOOK3S_64
716void slice_setup_new_exec(void)
717{
718 struct mm_struct *mm = current->mm;
719
720 slice_dbg("slice_setup_new_exec(mm=%p)\n", mm);
721
722 if (!is_32bit_task())
723 return;
724
725 mm_ctx_set_slb_addr_limit(&mm->context, DEFAULT_MAP_WINDOW);
726}
727#endif
728
729void slice_set_range_psize(struct mm_struct *mm, unsigned long start,
730 unsigned long len, unsigned int psize)
731{
732 struct slice_mask mask;
733
734 VM_BUG_ON(radix_enabled());
735
736 slice_range_to_mask(start, len, &mask);
737 slice_convert(mm, &mask, psize);
738}
739
740#ifdef CONFIG_HUGETLB_PAGE
741/*
742 * is_hugepage_only_range() is used by generic code to verify whether
743 * a normal mmap mapping (non hugetlbfs) is valid on a given area.
744 *
745 * until the generic code provides a more generic hook and/or starts
746 * calling arch get_unmapped_area for MAP_FIXED (which our implementation
747 * here knows how to deal with), we hijack it to keep standard mappings
748 * away from us.
749 *
750 * because of that generic code limitation, MAP_FIXED mapping cannot
751 * "convert" back a slice with no VMAs to the standard page size, only
752 * get_unmapped_area() can. It would be possible to fix it here but I
753 * prefer working on fixing the generic code instead.
754 *
755 * WARNING: This will not work if hugetlbfs isn't enabled since the
756 * generic code will redefine that function as 0 in that. This is ok
757 * for now as we only use slices with hugetlbfs enabled. This should
758 * be fixed as the generic code gets fixed.
759 */
760int slice_is_hugepage_only_range(struct mm_struct *mm, unsigned long addr,
761 unsigned long len)
762{
763 const struct slice_mask *maskp;
764 unsigned int psize = mm_ctx_user_psize(&mm->context);
765
766 VM_BUG_ON(radix_enabled());
767
768 maskp = slice_mask_for_size(&mm->context, psize);
769
770 /* We need to account for 4k slices too */
771 if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && psize == MMU_PAGE_64K) {
772 const struct slice_mask *compat_maskp;
773 struct slice_mask available;
774
775 compat_maskp = slice_mask_for_size(&mm->context, MMU_PAGE_4K);
776 slice_or_mask(&available, maskp, compat_maskp);
777 return !slice_check_range_fits(mm, &available, addr, len);
778 }
779
780 return !slice_check_range_fits(mm, maskp, addr, len);
781}
782#endif
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/module.h>
33#include <asm/mman.h>
34#include <asm/mmu.h>
35#include <asm/spu.h>
36
37static DEFINE_SPINLOCK(slice_convert_lock);
38
39
40#ifdef DEBUG
41int _slice_debug = 1;
42
43static void slice_print_mask(const char *label, struct slice_mask mask)
44{
45 char *p, buf[16 + 3 + 16 + 1];
46 int i;
47
48 if (!_slice_debug)
49 return;
50 p = buf;
51 for (i = 0; i < SLICE_NUM_LOW; i++)
52 *(p++) = (mask.low_slices & (1 << i)) ? '1' : '0';
53 *(p++) = ' ';
54 *(p++) = '-';
55 *(p++) = ' ';
56 for (i = 0; i < SLICE_NUM_HIGH; i++)
57 *(p++) = (mask.high_slices & (1 << i)) ? '1' : '0';
58 *(p++) = 0;
59
60 printk(KERN_DEBUG "%s:%s\n", label, buf);
61}
62
63#define slice_dbg(fmt...) do { if (_slice_debug) pr_debug(fmt); } while(0)
64
65#else
66
67static void slice_print_mask(const char *label, struct slice_mask mask) {}
68#define slice_dbg(fmt...)
69
70#endif
71
72static struct slice_mask slice_range_to_mask(unsigned long start,
73 unsigned long len)
74{
75 unsigned long end = start + len - 1;
76 struct slice_mask ret = { 0, 0 };
77
78 if (start < SLICE_LOW_TOP) {
79 unsigned long mend = min(end, SLICE_LOW_TOP);
80 unsigned long mstart = min(start, SLICE_LOW_TOP);
81
82 ret.low_slices = (1u << (GET_LOW_SLICE_INDEX(mend) + 1))
83 - (1u << GET_LOW_SLICE_INDEX(mstart));
84 }
85
86 if ((start + len) > SLICE_LOW_TOP)
87 ret.high_slices = (1u << (GET_HIGH_SLICE_INDEX(end) + 1))
88 - (1u << GET_HIGH_SLICE_INDEX(start));
89
90 return ret;
91}
92
93static int slice_area_is_free(struct mm_struct *mm, unsigned long addr,
94 unsigned long len)
95{
96 struct vm_area_struct *vma;
97
98 if ((mm->task_size - len) < addr)
99 return 0;
100 vma = find_vma(mm, addr);
101 return (!vma || (addr + len) <= vma->vm_start);
102}
103
104static int slice_low_has_vma(struct mm_struct *mm, unsigned long slice)
105{
106 return !slice_area_is_free(mm, slice << SLICE_LOW_SHIFT,
107 1ul << SLICE_LOW_SHIFT);
108}
109
110static int slice_high_has_vma(struct mm_struct *mm, unsigned long slice)
111{
112 unsigned long start = slice << SLICE_HIGH_SHIFT;
113 unsigned long end = start + (1ul << SLICE_HIGH_SHIFT);
114
115 /* Hack, so that each addresses is controlled by exactly one
116 * of the high or low area bitmaps, the first high area starts
117 * at 4GB, not 0 */
118 if (start == 0)
119 start = SLICE_LOW_TOP;
120
121 return !slice_area_is_free(mm, start, end - start);
122}
123
124static struct slice_mask slice_mask_for_free(struct mm_struct *mm)
125{
126 struct slice_mask ret = { 0, 0 };
127 unsigned long i;
128
129 for (i = 0; i < SLICE_NUM_LOW; i++)
130 if (!slice_low_has_vma(mm, i))
131 ret.low_slices |= 1u << i;
132
133 if (mm->task_size <= SLICE_LOW_TOP)
134 return ret;
135
136 for (i = 0; i < SLICE_NUM_HIGH; i++)
137 if (!slice_high_has_vma(mm, i))
138 ret.high_slices |= 1u << i;
139
140 return ret;
141}
142
143static struct slice_mask slice_mask_for_size(struct mm_struct *mm, int psize)
144{
145 struct slice_mask ret = { 0, 0 };
146 unsigned long i;
147 u64 psizes;
148
149 psizes = mm->context.low_slices_psize;
150 for (i = 0; i < SLICE_NUM_LOW; i++)
151 if (((psizes >> (i * 4)) & 0xf) == psize)
152 ret.low_slices |= 1u << i;
153
154 psizes = mm->context.high_slices_psize;
155 for (i = 0; i < SLICE_NUM_HIGH; i++)
156 if (((psizes >> (i * 4)) & 0xf) == psize)
157 ret.high_slices |= 1u << i;
158
159 return ret;
160}
161
162static int slice_check_fit(struct slice_mask mask, struct slice_mask available)
163{
164 return (mask.low_slices & available.low_slices) == mask.low_slices &&
165 (mask.high_slices & available.high_slices) == mask.high_slices;
166}
167
168static void slice_flush_segments(void *parm)
169{
170 struct mm_struct *mm = parm;
171 unsigned long flags;
172
173 if (mm != current->active_mm)
174 return;
175
176 /* update the paca copy of the context struct */
177 get_paca()->context = current->active_mm->context;
178
179 local_irq_save(flags);
180 slb_flush_and_rebolt();
181 local_irq_restore(flags);
182}
183
184static void slice_convert(struct mm_struct *mm, struct slice_mask mask, int psize)
185{
186 /* Write the new slice psize bits */
187 u64 lpsizes, hpsizes;
188 unsigned long i, flags;
189
190 slice_dbg("slice_convert(mm=%p, psize=%d)\n", mm, psize);
191 slice_print_mask(" mask", mask);
192
193 /* We need to use a spinlock here to protect against
194 * concurrent 64k -> 4k demotion ...
195 */
196 spin_lock_irqsave(&slice_convert_lock, flags);
197
198 lpsizes = mm->context.low_slices_psize;
199 for (i = 0; i < SLICE_NUM_LOW; i++)
200 if (mask.low_slices & (1u << i))
201 lpsizes = (lpsizes & ~(0xful << (i * 4))) |
202 (((unsigned long)psize) << (i * 4));
203
204 hpsizes = mm->context.high_slices_psize;
205 for (i = 0; i < SLICE_NUM_HIGH; i++)
206 if (mask.high_slices & (1u << i))
207 hpsizes = (hpsizes & ~(0xful << (i * 4))) |
208 (((unsigned long)psize) << (i * 4));
209
210 mm->context.low_slices_psize = lpsizes;
211 mm->context.high_slices_psize = hpsizes;
212
213 slice_dbg(" lsps=%lx, hsps=%lx\n",
214 mm->context.low_slices_psize,
215 mm->context.high_slices_psize);
216
217 spin_unlock_irqrestore(&slice_convert_lock, flags);
218
219#ifdef CONFIG_SPU_BASE
220 spu_flush_all_slbs(mm);
221#endif
222}
223
224static unsigned long slice_find_area_bottomup(struct mm_struct *mm,
225 unsigned long len,
226 struct slice_mask available,
227 int psize, int use_cache)
228{
229 struct vm_area_struct *vma;
230 unsigned long start_addr, addr;
231 struct slice_mask mask;
232 int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
233
234 if (use_cache) {
235 if (len <= mm->cached_hole_size) {
236 start_addr = addr = TASK_UNMAPPED_BASE;
237 mm->cached_hole_size = 0;
238 } else
239 start_addr = addr = mm->free_area_cache;
240 } else
241 start_addr = addr = TASK_UNMAPPED_BASE;
242
243full_search:
244 for (;;) {
245 addr = _ALIGN_UP(addr, 1ul << pshift);
246 if ((TASK_SIZE - len) < addr)
247 break;
248 vma = find_vma(mm, addr);
249 BUG_ON(vma && (addr >= vma->vm_end));
250
251 mask = slice_range_to_mask(addr, len);
252 if (!slice_check_fit(mask, available)) {
253 if (addr < SLICE_LOW_TOP)
254 addr = _ALIGN_UP(addr + 1, 1ul << SLICE_LOW_SHIFT);
255 else
256 addr = _ALIGN_UP(addr + 1, 1ul << SLICE_HIGH_SHIFT);
257 continue;
258 }
259 if (!vma || addr + len <= vma->vm_start) {
260 /*
261 * Remember the place where we stopped the search:
262 */
263 if (use_cache)
264 mm->free_area_cache = addr + len;
265 return addr;
266 }
267 if (use_cache && (addr + mm->cached_hole_size) < vma->vm_start)
268 mm->cached_hole_size = vma->vm_start - addr;
269 addr = vma->vm_end;
270 }
271
272 /* Make sure we didn't miss any holes */
273 if (use_cache && start_addr != TASK_UNMAPPED_BASE) {
274 start_addr = addr = TASK_UNMAPPED_BASE;
275 mm->cached_hole_size = 0;
276 goto full_search;
277 }
278 return -ENOMEM;
279}
280
281static unsigned long slice_find_area_topdown(struct mm_struct *mm,
282 unsigned long len,
283 struct slice_mask available,
284 int psize, int use_cache)
285{
286 struct vm_area_struct *vma;
287 unsigned long addr;
288 struct slice_mask mask;
289 int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
290
291 /* check if free_area_cache is useful for us */
292 if (use_cache) {
293 if (len <= mm->cached_hole_size) {
294 mm->cached_hole_size = 0;
295 mm->free_area_cache = mm->mmap_base;
296 }
297
298 /* either no address requested or can't fit in requested
299 * address hole
300 */
301 addr = mm->free_area_cache;
302
303 /* make sure it can fit in the remaining address space */
304 if (addr > len) {
305 addr = _ALIGN_DOWN(addr - len, 1ul << pshift);
306 mask = slice_range_to_mask(addr, len);
307 if (slice_check_fit(mask, available) &&
308 slice_area_is_free(mm, addr, len))
309 /* remember the address as a hint for
310 * next time
311 */
312 return (mm->free_area_cache = addr);
313 }
314 }
315
316 addr = mm->mmap_base;
317 while (addr > len) {
318 /* Go down by chunk size */
319 addr = _ALIGN_DOWN(addr - len, 1ul << pshift);
320
321 /* Check for hit with different page size */
322 mask = slice_range_to_mask(addr, len);
323 if (!slice_check_fit(mask, available)) {
324 if (addr < SLICE_LOW_TOP)
325 addr = _ALIGN_DOWN(addr, 1ul << SLICE_LOW_SHIFT);
326 else if (addr < (1ul << SLICE_HIGH_SHIFT))
327 addr = SLICE_LOW_TOP;
328 else
329 addr = _ALIGN_DOWN(addr, 1ul << SLICE_HIGH_SHIFT);
330 continue;
331 }
332
333 /*
334 * Lookup failure means no vma is above this address,
335 * else if new region fits below vma->vm_start,
336 * return with success:
337 */
338 vma = find_vma(mm, addr);
339 if (!vma || (addr + len) <= vma->vm_start) {
340 /* remember the address as a hint for next time */
341 if (use_cache)
342 mm->free_area_cache = addr;
343 return addr;
344 }
345
346 /* remember the largest hole we saw so far */
347 if (use_cache && (addr + mm->cached_hole_size) < vma->vm_start)
348 mm->cached_hole_size = vma->vm_start - addr;
349
350 /* try just below the current vma->vm_start */
351 addr = vma->vm_start;
352 }
353
354 /*
355 * A failed mmap() very likely causes application failure,
356 * so fall back to the bottom-up function here. This scenario
357 * can happen with large stack limits and large mmap()
358 * allocations.
359 */
360 addr = slice_find_area_bottomup(mm, len, available, psize, 0);
361
362 /*
363 * Restore the topdown base:
364 */
365 if (use_cache) {
366 mm->free_area_cache = mm->mmap_base;
367 mm->cached_hole_size = ~0UL;
368 }
369
370 return addr;
371}
372
373
374static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len,
375 struct slice_mask mask, int psize,
376 int topdown, int use_cache)
377{
378 if (topdown)
379 return slice_find_area_topdown(mm, len, mask, psize, use_cache);
380 else
381 return slice_find_area_bottomup(mm, len, mask, psize, use_cache);
382}
383
384#define or_mask(dst, src) do { \
385 (dst).low_slices |= (src).low_slices; \
386 (dst).high_slices |= (src).high_slices; \
387} while (0)
388
389#define andnot_mask(dst, src) do { \
390 (dst).low_slices &= ~(src).low_slices; \
391 (dst).high_slices &= ~(src).high_slices; \
392} while (0)
393
394#ifdef CONFIG_PPC_64K_PAGES
395#define MMU_PAGE_BASE MMU_PAGE_64K
396#else
397#define MMU_PAGE_BASE MMU_PAGE_4K
398#endif
399
400unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
401 unsigned long flags, unsigned int psize,
402 int topdown, int use_cache)
403{
404 struct slice_mask mask = {0, 0};
405 struct slice_mask good_mask;
406 struct slice_mask potential_mask = {0,0} /* silence stupid warning */;
407 struct slice_mask compat_mask = {0, 0};
408 int fixed = (flags & MAP_FIXED);
409 int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
410 struct mm_struct *mm = current->mm;
411 unsigned long newaddr;
412
413 /* Sanity checks */
414 BUG_ON(mm->task_size == 0);
415
416 slice_dbg("slice_get_unmapped_area(mm=%p, psize=%d...\n", mm, psize);
417 slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d, use_cache=%d\n",
418 addr, len, flags, topdown, use_cache);
419
420 if (len > mm->task_size)
421 return -ENOMEM;
422 if (len & ((1ul << pshift) - 1))
423 return -EINVAL;
424 if (fixed && (addr & ((1ul << pshift) - 1)))
425 return -EINVAL;
426 if (fixed && addr > (mm->task_size - len))
427 return -EINVAL;
428
429 /* If hint, make sure it matches our alignment restrictions */
430 if (!fixed && addr) {
431 addr = _ALIGN_UP(addr, 1ul << pshift);
432 slice_dbg(" aligned addr=%lx\n", addr);
433 /* Ignore hint if it's too large or overlaps a VMA */
434 if (addr > mm->task_size - len ||
435 !slice_area_is_free(mm, addr, len))
436 addr = 0;
437 }
438
439 /* First make up a "good" mask of slices that have the right size
440 * already
441 */
442 good_mask = slice_mask_for_size(mm, psize);
443 slice_print_mask(" good_mask", good_mask);
444
445 /*
446 * Here "good" means slices that are already the right page size,
447 * "compat" means slices that have a compatible page size (i.e.
448 * 4k in a 64k pagesize kernel), and "free" means slices without
449 * any VMAs.
450 *
451 * If MAP_FIXED:
452 * check if fits in good | compat => OK
453 * check if fits in good | compat | free => convert free
454 * else bad
455 * If have hint:
456 * check if hint fits in good => OK
457 * check if hint fits in good | free => convert free
458 * Otherwise:
459 * search in good, found => OK
460 * search in good | free, found => convert free
461 * search in good | compat | free, found => convert free.
462 */
463
464#ifdef CONFIG_PPC_64K_PAGES
465 /* If we support combo pages, we can allow 64k pages in 4k slices */
466 if (psize == MMU_PAGE_64K) {
467 compat_mask = slice_mask_for_size(mm, MMU_PAGE_4K);
468 if (fixed)
469 or_mask(good_mask, compat_mask);
470 }
471#endif
472
473 /* First check hint if it's valid or if we have MAP_FIXED */
474 if (addr != 0 || fixed) {
475 /* Build a mask for the requested range */
476 mask = slice_range_to_mask(addr, len);
477 slice_print_mask(" mask", mask);
478
479 /* Check if we fit in the good mask. If we do, we just return,
480 * nothing else to do
481 */
482 if (slice_check_fit(mask, good_mask)) {
483 slice_dbg(" fits good !\n");
484 return addr;
485 }
486 } else {
487 /* Now let's see if we can find something in the existing
488 * slices for that size
489 */
490 newaddr = slice_find_area(mm, len, good_mask, psize, topdown,
491 use_cache);
492 if (newaddr != -ENOMEM) {
493 /* Found within the good mask, we don't have to setup,
494 * we thus return directly
495 */
496 slice_dbg(" found area at 0x%lx\n", newaddr);
497 return newaddr;
498 }
499 }
500
501 /* We don't fit in the good mask, check what other slices are
502 * empty and thus can be converted
503 */
504 potential_mask = slice_mask_for_free(mm);
505 or_mask(potential_mask, good_mask);
506 slice_print_mask(" potential", potential_mask);
507
508 if ((addr != 0 || fixed) && slice_check_fit(mask, potential_mask)) {
509 slice_dbg(" fits potential !\n");
510 goto convert;
511 }
512
513 /* If we have MAP_FIXED and failed the above steps, then error out */
514 if (fixed)
515 return -EBUSY;
516
517 slice_dbg(" search...\n");
518
519 /* If we had a hint that didn't work out, see if we can fit
520 * anywhere in the good area.
521 */
522 if (addr) {
523 addr = slice_find_area(mm, len, good_mask, psize, topdown,
524 use_cache);
525 if (addr != -ENOMEM) {
526 slice_dbg(" found area at 0x%lx\n", addr);
527 return addr;
528 }
529 }
530
531 /* Now let's see if we can find something in the existing slices
532 * for that size plus free slices
533 */
534 addr = slice_find_area(mm, len, potential_mask, psize, topdown,
535 use_cache);
536
537#ifdef CONFIG_PPC_64K_PAGES
538 if (addr == -ENOMEM && psize == MMU_PAGE_64K) {
539 /* retry the search with 4k-page slices included */
540 or_mask(potential_mask, compat_mask);
541 addr = slice_find_area(mm, len, potential_mask, psize,
542 topdown, use_cache);
543 }
544#endif
545
546 if (addr == -ENOMEM)
547 return -ENOMEM;
548
549 mask = slice_range_to_mask(addr, len);
550 slice_dbg(" found potential area at 0x%lx\n", addr);
551 slice_print_mask(" mask", mask);
552
553 convert:
554 andnot_mask(mask, good_mask);
555 andnot_mask(mask, compat_mask);
556 if (mask.low_slices || mask.high_slices) {
557 slice_convert(mm, mask, psize);
558 if (psize > MMU_PAGE_BASE)
559 on_each_cpu(slice_flush_segments, mm, 1);
560 }
561 return addr;
562
563}
564EXPORT_SYMBOL_GPL(slice_get_unmapped_area);
565
566unsigned long arch_get_unmapped_area(struct file *filp,
567 unsigned long addr,
568 unsigned long len,
569 unsigned long pgoff,
570 unsigned long flags)
571{
572 return slice_get_unmapped_area(addr, len, flags,
573 current->mm->context.user_psize,
574 0, 1);
575}
576
577unsigned long arch_get_unmapped_area_topdown(struct file *filp,
578 const unsigned long addr0,
579 const unsigned long len,
580 const unsigned long pgoff,
581 const unsigned long flags)
582{
583 return slice_get_unmapped_area(addr0, len, flags,
584 current->mm->context.user_psize,
585 1, 1);
586}
587
588unsigned int get_slice_psize(struct mm_struct *mm, unsigned long addr)
589{
590 u64 psizes;
591 int index;
592
593 if (addr < SLICE_LOW_TOP) {
594 psizes = mm->context.low_slices_psize;
595 index = GET_LOW_SLICE_INDEX(addr);
596 } else {
597 psizes = mm->context.high_slices_psize;
598 index = GET_HIGH_SLICE_INDEX(addr);
599 }
600
601 return (psizes >> (index * 4)) & 0xf;
602}
603EXPORT_SYMBOL_GPL(get_slice_psize);
604
605/*
606 * This is called by hash_page when it needs to do a lazy conversion of
607 * an address space from real 64K pages to combo 4K pages (typically
608 * when hitting a non cacheable mapping on a processor or hypervisor
609 * that won't allow them for 64K pages).
610 *
611 * This is also called in init_new_context() to change back the user
612 * psize from whatever the parent context had it set to
613 * N.B. This may be called before mm->context.id has been set.
614 *
615 * This function will only change the content of the {low,high)_slice_psize
616 * masks, it will not flush SLBs as this shall be handled lazily by the
617 * caller.
618 */
619void slice_set_user_psize(struct mm_struct *mm, unsigned int psize)
620{
621 unsigned long flags, lpsizes, hpsizes;
622 unsigned int old_psize;
623 int i;
624
625 slice_dbg("slice_set_user_psize(mm=%p, psize=%d)\n", mm, psize);
626
627 spin_lock_irqsave(&slice_convert_lock, flags);
628
629 old_psize = mm->context.user_psize;
630 slice_dbg(" old_psize=%d\n", old_psize);
631 if (old_psize == psize)
632 goto bail;
633
634 mm->context.user_psize = psize;
635 wmb();
636
637 lpsizes = mm->context.low_slices_psize;
638 for (i = 0; i < SLICE_NUM_LOW; i++)
639 if (((lpsizes >> (i * 4)) & 0xf) == old_psize)
640 lpsizes = (lpsizes & ~(0xful << (i * 4))) |
641 (((unsigned long)psize) << (i * 4));
642
643 hpsizes = mm->context.high_slices_psize;
644 for (i = 0; i < SLICE_NUM_HIGH; i++)
645 if (((hpsizes >> (i * 4)) & 0xf) == old_psize)
646 hpsizes = (hpsizes & ~(0xful << (i * 4))) |
647 (((unsigned long)psize) << (i * 4));
648
649 mm->context.low_slices_psize = lpsizes;
650 mm->context.high_slices_psize = hpsizes;
651
652 slice_dbg(" lsps=%lx, hsps=%lx\n",
653 mm->context.low_slices_psize,
654 mm->context.high_slices_psize);
655
656 bail:
657 spin_unlock_irqrestore(&slice_convert_lock, flags);
658}
659
660void slice_set_psize(struct mm_struct *mm, unsigned long address,
661 unsigned int psize)
662{
663 unsigned long i, flags;
664 u64 *p;
665
666 spin_lock_irqsave(&slice_convert_lock, flags);
667 if (address < SLICE_LOW_TOP) {
668 i = GET_LOW_SLICE_INDEX(address);
669 p = &mm->context.low_slices_psize;
670 } else {
671 i = GET_HIGH_SLICE_INDEX(address);
672 p = &mm->context.high_slices_psize;
673 }
674 *p = (*p & ~(0xful << (i * 4))) | ((unsigned long) psize << (i * 4));
675 spin_unlock_irqrestore(&slice_convert_lock, flags);
676
677#ifdef CONFIG_SPU_BASE
678 spu_flush_all_slbs(mm);
679#endif
680}
681
682void slice_set_range_psize(struct mm_struct *mm, unsigned long start,
683 unsigned long len, unsigned int psize)
684{
685 struct slice_mask mask = slice_range_to_mask(start, len);
686
687 slice_convert(mm, mask, psize);
688}
689
690/*
691 * is_hugepage_only_range() is used by generic code to verify wether
692 * a normal mmap mapping (non hugetlbfs) is valid on a given area.
693 *
694 * until the generic code provides a more generic hook and/or starts
695 * calling arch get_unmapped_area for MAP_FIXED (which our implementation
696 * here knows how to deal with), we hijack it to keep standard mappings
697 * away from us.
698 *
699 * because of that generic code limitation, MAP_FIXED mapping cannot
700 * "convert" back a slice with no VMAs to the standard page size, only
701 * get_unmapped_area() can. It would be possible to fix it here but I
702 * prefer working on fixing the generic code instead.
703 *
704 * WARNING: This will not work if hugetlbfs isn't enabled since the
705 * generic code will redefine that function as 0 in that. This is ok
706 * for now as we only use slices with hugetlbfs enabled. This should
707 * be fixed as the generic code gets fixed.
708 */
709int is_hugepage_only_range(struct mm_struct *mm, unsigned long addr,
710 unsigned long len)
711{
712 struct slice_mask mask, available;
713 unsigned int psize = mm->context.user_psize;
714
715 mask = slice_range_to_mask(addr, len);
716 available = slice_mask_for_size(mm, psize);
717#ifdef CONFIG_PPC_64K_PAGES
718 /* We need to account for 4k slices too */
719 if (psize == MMU_PAGE_64K) {
720 struct slice_mask compat_mask;
721 compat_mask = slice_mask_for_size(mm, MMU_PAGE_4K);
722 or_mask(available, compat_mask);
723 }
724#endif
725
726#if 0 /* too verbose */
727 slice_dbg("is_hugepage_only_range(mm=%p, addr=%lx, len=%lx)\n",
728 mm, addr, len);
729 slice_print_mask(" mask", mask);
730 slice_print_mask(" available", available);
731#endif
732 return !slice_check_fit(mask, available);
733}
734