xref: /openbmc/linux/arch/powerpc/mm/book3s64/slice.c (revision 9a29f5fc)
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 
29 static DEFINE_SPINLOCK(slice_convert_lock);
30 
31 #ifdef DEBUG
32 int _slice_debug = 1;
33 
34 static 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 
48 static void slice_print_mask(const char *label, const struct slice_mask *mask) {}
49 #define slice_dbg(fmt...)
50 
51 #endif
52 
53 static inline notrace bool slice_addr_is_low(unsigned long addr)
54 {
55 	u64 tmp = (u64)addr;
56 
57 	return tmp < SLICE_LOW_TOP;
58 }
59 
60 static 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 
86 static 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 
97 static 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 
103 static 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 
117 static 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 
138 static 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 
170 static 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 
187 static 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  */
261 static 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 
278 static unsigned long slice_find_area_bottomup(struct mm_struct *mm,
279 					      unsigned long addr, 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 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 	 * Check till the allow max value for this mmap request
293 	 */
294 	while (addr < high_limit) {
295 		info.low_limit = addr;
296 		if (!slice_scan_available(addr, available, 1, &addr))
297 			continue;
298 
299  next_slice:
300 		/*
301 		 * At this point [info.low_limit; addr) covers
302 		 * available slices only and ends at a slice boundary.
303 		 * Check if we need to reduce the range, or if we can
304 		 * extend it to cover the next available slice.
305 		 */
306 		if (addr >= high_limit)
307 			addr = high_limit;
308 		else if (slice_scan_available(addr, available, 1, &next_end)) {
309 			addr = next_end;
310 			goto next_slice;
311 		}
312 		info.high_limit = addr;
313 
314 		found = vm_unmapped_area(&info);
315 		if (!(found & ~PAGE_MASK))
316 			return found;
317 	}
318 
319 	return -ENOMEM;
320 }
321 
322 static unsigned long slice_find_area_topdown(struct mm_struct *mm,
323 					     unsigned long addr, unsigned long len,
324 					     const struct slice_mask *available,
325 					     int psize, unsigned long high_limit)
326 {
327 	int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
328 	unsigned long found, prev;
329 	struct vm_unmapped_area_info info;
330 	unsigned long min_addr = max(PAGE_SIZE, mmap_min_addr);
331 
332 	info.flags = VM_UNMAPPED_AREA_TOPDOWN;
333 	info.length = len;
334 	info.align_mask = PAGE_MASK & ((1ul << pshift) - 1);
335 	info.align_offset = 0;
336 	/*
337 	 * If we are trying to allocate above DEFAULT_MAP_WINDOW
338 	 * Add the different to the mmap_base.
339 	 * Only for that request for which high_limit is above
340 	 * DEFAULT_MAP_WINDOW we should apply this.
341 	 */
342 	if (high_limit > DEFAULT_MAP_WINDOW)
343 		addr += mm_ctx_slb_addr_limit(&mm->context) - DEFAULT_MAP_WINDOW;
344 
345 	while (addr > min_addr) {
346 		info.high_limit = addr;
347 		if (!slice_scan_available(addr - 1, available, 0, &addr))
348 			continue;
349 
350  prev_slice:
351 		/*
352 		 * At this point [addr; info.high_limit) covers
353 		 * available slices only and starts at a slice boundary.
354 		 * Check if we need to reduce the range, or if we can
355 		 * extend it to cover the previous available slice.
356 		 */
357 		if (addr < min_addr)
358 			addr = min_addr;
359 		else if (slice_scan_available(addr - 1, available, 0, &prev)) {
360 			addr = prev;
361 			goto prev_slice;
362 		}
363 		info.low_limit = addr;
364 
365 		found = vm_unmapped_area(&info);
366 		if (!(found & ~PAGE_MASK))
367 			return found;
368 	}
369 
370 	/*
371 	 * A failed mmap() very likely causes application failure,
372 	 * so fall back to the bottom-up function here. This scenario
373 	 * can happen with large stack limits and large mmap()
374 	 * allocations.
375 	 */
376 	return slice_find_area_bottomup(mm, TASK_UNMAPPED_BASE, len, available, psize, high_limit);
377 }
378 
379 
380 static unsigned long slice_find_area(struct mm_struct *mm, unsigned long len,
381 				     const struct slice_mask *mask, int psize,
382 				     int topdown, unsigned long high_limit)
383 {
384 	if (topdown)
385 		return slice_find_area_topdown(mm, mm->mmap_base, len, mask, psize, high_limit);
386 	else
387 		return slice_find_area_bottomup(mm, mm->mmap_base, len, mask, psize, high_limit);
388 }
389 
390 static inline void slice_copy_mask(struct slice_mask *dst,
391 					const struct slice_mask *src)
392 {
393 	dst->low_slices = src->low_slices;
394 	if (!SLICE_NUM_HIGH)
395 		return;
396 	bitmap_copy(dst->high_slices, src->high_slices, SLICE_NUM_HIGH);
397 }
398 
399 static inline void slice_or_mask(struct slice_mask *dst,
400 					const struct slice_mask *src1,
401 					const struct slice_mask *src2)
402 {
403 	dst->low_slices = src1->low_slices | src2->low_slices;
404 	if (!SLICE_NUM_HIGH)
405 		return;
406 	bitmap_or(dst->high_slices, src1->high_slices, src2->high_slices, SLICE_NUM_HIGH);
407 }
408 
409 static inline void slice_andnot_mask(struct slice_mask *dst,
410 					const struct slice_mask *src1,
411 					const struct slice_mask *src2)
412 {
413 	dst->low_slices = src1->low_slices & ~src2->low_slices;
414 	if (!SLICE_NUM_HIGH)
415 		return;
416 	bitmap_andnot(dst->high_slices, src1->high_slices, src2->high_slices, SLICE_NUM_HIGH);
417 }
418 
419 #ifdef CONFIG_PPC_64K_PAGES
420 #define MMU_PAGE_BASE	MMU_PAGE_64K
421 #else
422 #define MMU_PAGE_BASE	MMU_PAGE_4K
423 #endif
424 
425 unsigned long slice_get_unmapped_area(unsigned long addr, unsigned long len,
426 				      unsigned long flags, unsigned int psize,
427 				      int topdown)
428 {
429 	struct slice_mask good_mask;
430 	struct slice_mask potential_mask;
431 	const struct slice_mask *maskp;
432 	const struct slice_mask *compat_maskp = NULL;
433 	int fixed = (flags & MAP_FIXED);
434 	int pshift = max_t(int, mmu_psize_defs[psize].shift, PAGE_SHIFT);
435 	unsigned long page_size = 1UL << pshift;
436 	struct mm_struct *mm = current->mm;
437 	unsigned long newaddr;
438 	unsigned long high_limit;
439 
440 	high_limit = DEFAULT_MAP_WINDOW;
441 	if (addr >= high_limit || (fixed && (addr + len > high_limit)))
442 		high_limit = TASK_SIZE;
443 
444 	if (len > high_limit)
445 		return -ENOMEM;
446 	if (len & (page_size - 1))
447 		return -EINVAL;
448 	if (fixed) {
449 		if (addr & (page_size - 1))
450 			return -EINVAL;
451 		if (addr > high_limit - len)
452 			return -ENOMEM;
453 	}
454 
455 	if (high_limit > mm_ctx_slb_addr_limit(&mm->context)) {
456 		/*
457 		 * Increasing the slb_addr_limit does not require
458 		 * slice mask cache to be recalculated because it should
459 		 * be already initialised beyond the old address limit.
460 		 */
461 		mm_ctx_set_slb_addr_limit(&mm->context, high_limit);
462 
463 		on_each_cpu(slice_flush_segments, mm, 1);
464 	}
465 
466 	/* Sanity checks */
467 	BUG_ON(mm->task_size == 0);
468 	BUG_ON(mm_ctx_slb_addr_limit(&mm->context) == 0);
469 	VM_BUG_ON(radix_enabled());
470 
471 	slice_dbg("slice_get_unmapped_area(mm=%p, psize=%d...\n", mm, psize);
472 	slice_dbg(" addr=%lx, len=%lx, flags=%lx, topdown=%d\n",
473 		  addr, len, flags, topdown);
474 
475 	/* If hint, make sure it matches our alignment restrictions */
476 	if (!fixed && addr) {
477 		addr = ALIGN(addr, page_size);
478 		slice_dbg(" aligned addr=%lx\n", addr);
479 		/* Ignore hint if it's too large or overlaps a VMA */
480 		if (addr > high_limit - len || addr < mmap_min_addr ||
481 		    !slice_area_is_free(mm, addr, len))
482 			addr = 0;
483 	}
484 
485 	/* First make up a "good" mask of slices that have the right size
486 	 * already
487 	 */
488 	maskp = slice_mask_for_size(&mm->context, psize);
489 
490 	/*
491 	 * Here "good" means slices that are already the right page size,
492 	 * "compat" means slices that have a compatible page size (i.e.
493 	 * 4k in a 64k pagesize kernel), and "free" means slices without
494 	 * any VMAs.
495 	 *
496 	 * If MAP_FIXED:
497 	 *	check if fits in good | compat => OK
498 	 *	check if fits in good | compat | free => convert free
499 	 *	else bad
500 	 * If have hint:
501 	 *	check if hint fits in good => OK
502 	 *	check if hint fits in good | free => convert free
503 	 * Otherwise:
504 	 *	search in good, found => OK
505 	 *	search in good | free, found => convert free
506 	 *	search in good | compat | free, found => convert free.
507 	 */
508 
509 	/*
510 	 * If we support combo pages, we can allow 64k pages in 4k slices
511 	 * The mask copies could be avoided in most cases here if we had
512 	 * a pointer to good mask for the next code to use.
513 	 */
514 	if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && psize == MMU_PAGE_64K) {
515 		compat_maskp = slice_mask_for_size(&mm->context, MMU_PAGE_4K);
516 		if (fixed)
517 			slice_or_mask(&good_mask, maskp, compat_maskp);
518 		else
519 			slice_copy_mask(&good_mask, maskp);
520 	} else {
521 		slice_copy_mask(&good_mask, maskp);
522 	}
523 
524 	slice_print_mask(" good_mask", &good_mask);
525 	if (compat_maskp)
526 		slice_print_mask(" compat_mask", compat_maskp);
527 
528 	/* First check hint if it's valid or if we have MAP_FIXED */
529 	if (addr != 0 || fixed) {
530 		/* Check if we fit in the good mask. If we do, we just return,
531 		 * nothing else to do
532 		 */
533 		if (slice_check_range_fits(mm, &good_mask, addr, len)) {
534 			slice_dbg(" fits good !\n");
535 			newaddr = addr;
536 			goto return_addr;
537 		}
538 	} else {
539 		/* Now let's see if we can find something in the existing
540 		 * slices for that size
541 		 */
542 		newaddr = slice_find_area(mm, len, &good_mask,
543 					  psize, topdown, high_limit);
544 		if (newaddr != -ENOMEM) {
545 			/* Found within the good mask, we don't have to setup,
546 			 * we thus return directly
547 			 */
548 			slice_dbg(" found area at 0x%lx\n", newaddr);
549 			goto return_addr;
550 		}
551 	}
552 	/*
553 	 * We don't fit in the good mask, check what other slices are
554 	 * empty and thus can be converted
555 	 */
556 	slice_mask_for_free(mm, &potential_mask, high_limit);
557 	slice_or_mask(&potential_mask, &potential_mask, &good_mask);
558 	slice_print_mask(" potential", &potential_mask);
559 
560 	if (addr != 0 || fixed) {
561 		if (slice_check_range_fits(mm, &potential_mask, addr, len)) {
562 			slice_dbg(" fits potential !\n");
563 			newaddr = addr;
564 			goto convert;
565 		}
566 	}
567 
568 	/* If we have MAP_FIXED and failed the above steps, then error out */
569 	if (fixed)
570 		return -EBUSY;
571 
572 	slice_dbg(" search...\n");
573 
574 	/* If we had a hint that didn't work out, see if we can fit
575 	 * anywhere in the good area.
576 	 */
577 	if (addr) {
578 		newaddr = slice_find_area(mm, len, &good_mask,
579 					  psize, topdown, high_limit);
580 		if (newaddr != -ENOMEM) {
581 			slice_dbg(" found area at 0x%lx\n", newaddr);
582 			goto return_addr;
583 		}
584 	}
585 
586 	/* Now let's see if we can find something in the existing slices
587 	 * for that size plus free slices
588 	 */
589 	newaddr = slice_find_area(mm, len, &potential_mask,
590 				  psize, topdown, high_limit);
591 
592 	if (IS_ENABLED(CONFIG_PPC_64K_PAGES) && newaddr == -ENOMEM &&
593 	    psize == MMU_PAGE_64K) {
594 		/* retry the search with 4k-page slices included */
595 		slice_or_mask(&potential_mask, &potential_mask, compat_maskp);
596 		newaddr = slice_find_area(mm, len, &potential_mask,
597 					  psize, topdown, high_limit);
598 	}
599 
600 	if (newaddr == -ENOMEM)
601 		return -ENOMEM;
602 
603 	slice_range_to_mask(newaddr, len, &potential_mask);
604 	slice_dbg(" found potential area at 0x%lx\n", newaddr);
605 	slice_print_mask(" mask", &potential_mask);
606 
607  convert:
608 	/*
609 	 * Try to allocate the context before we do slice convert
610 	 * so that we handle the context allocation failure gracefully.
611 	 */
612 	if (need_extra_context(mm, newaddr)) {
613 		if (alloc_extended_context(mm, newaddr) < 0)
614 			return -ENOMEM;
615 	}
616 
617 	slice_andnot_mask(&potential_mask, &potential_mask, &good_mask);
618 	if (compat_maskp && !fixed)
619 		slice_andnot_mask(&potential_mask, &potential_mask, compat_maskp);
620 	if (potential_mask.low_slices ||
621 		(SLICE_NUM_HIGH &&
622 		 !bitmap_empty(potential_mask.high_slices, SLICE_NUM_HIGH))) {
623 		slice_convert(mm, &potential_mask, psize);
624 		if (psize > MMU_PAGE_BASE)
625 			on_each_cpu(slice_flush_segments, mm, 1);
626 	}
627 	return newaddr;
628 
629 return_addr:
630 	if (need_extra_context(mm, newaddr)) {
631 		if (alloc_extended_context(mm, newaddr) < 0)
632 			return -ENOMEM;
633 	}
634 	return newaddr;
635 }
636 EXPORT_SYMBOL_GPL(slice_get_unmapped_area);
637 
638 unsigned long arch_get_unmapped_area(struct file *filp,
639 				     unsigned long addr,
640 				     unsigned long len,
641 				     unsigned long pgoff,
642 				     unsigned long flags)
643 {
644 	if (radix_enabled())
645 		return generic_get_unmapped_area(filp, addr, len, pgoff, flags);
646 
647 	return slice_get_unmapped_area(addr, len, flags,
648 				       mm_ctx_user_psize(&current->mm->context), 0);
649 }
650 
651 unsigned long arch_get_unmapped_area_topdown(struct file *filp,
652 					     const unsigned long addr0,
653 					     const unsigned long len,
654 					     const unsigned long pgoff,
655 					     const unsigned long flags)
656 {
657 	if (radix_enabled())
658 		return generic_get_unmapped_area_topdown(filp, addr0, len, pgoff, flags);
659 
660 	return slice_get_unmapped_area(addr0, len, flags,
661 				       mm_ctx_user_psize(&current->mm->context), 1);
662 }
663 
664 unsigned int notrace get_slice_psize(struct mm_struct *mm, unsigned long addr)
665 {
666 	unsigned char *psizes;
667 	int index, mask_index;
668 
669 	VM_BUG_ON(radix_enabled());
670 
671 	if (slice_addr_is_low(addr)) {
672 		psizes = mm_ctx_low_slices(&mm->context);
673 		index = GET_LOW_SLICE_INDEX(addr);
674 	} else {
675 		psizes = mm_ctx_high_slices(&mm->context);
676 		index = GET_HIGH_SLICE_INDEX(addr);
677 	}
678 	mask_index = index & 0x1;
679 	return (psizes[index >> 1] >> (mask_index * 4)) & 0xf;
680 }
681 EXPORT_SYMBOL_GPL(get_slice_psize);
682 
683 void slice_init_new_context_exec(struct mm_struct *mm)
684 {
685 	unsigned char *hpsizes, *lpsizes;
686 	struct slice_mask *mask;
687 	unsigned int psize = mmu_virtual_psize;
688 
689 	slice_dbg("slice_init_new_context_exec(mm=%p)\n", mm);
690 
691 	/*
692 	 * In the case of exec, use the default limit. In the
693 	 * case of fork it is just inherited from the mm being
694 	 * duplicated.
695 	 */
696 	mm_ctx_set_slb_addr_limit(&mm->context, SLB_ADDR_LIMIT_DEFAULT);
697 	mm_ctx_set_user_psize(&mm->context, psize);
698 
699 	/*
700 	 * Set all slice psizes to the default.
701 	 */
702 	lpsizes = mm_ctx_low_slices(&mm->context);
703 	memset(lpsizes, (psize << 4) | psize, SLICE_NUM_LOW >> 1);
704 
705 	hpsizes = mm_ctx_high_slices(&mm->context);
706 	memset(hpsizes, (psize << 4) | psize, SLICE_NUM_HIGH >> 1);
707 
708 	/*
709 	 * Slice mask cache starts zeroed, fill the default size cache.
710 	 */
711 	mask = slice_mask_for_size(&mm->context, psize);
712 	mask->low_slices = ~0UL;
713 	if (SLICE_NUM_HIGH)
714 		bitmap_fill(mask->high_slices, SLICE_NUM_HIGH);
715 }
716 
717 void slice_setup_new_exec(void)
718 {
719 	struct mm_struct *mm = current->mm;
720 
721 	slice_dbg("slice_setup_new_exec(mm=%p)\n", mm);
722 
723 	if (!is_32bit_task())
724 		return;
725 
726 	mm_ctx_set_slb_addr_limit(&mm->context, DEFAULT_MAP_WINDOW);
727 }
728 
729 void 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  */
760 int 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 
783 unsigned long vma_mmu_pagesize(struct vm_area_struct *vma)
784 {
785 	/* With radix we don't use slice, so derive it from vma*/
786 	if (radix_enabled())
787 		return vma_kernel_pagesize(vma);
788 
789 	return 1UL << mmu_psize_to_shift(get_slice_psize(vma->vm_mm, vma->vm_start));
790 }
791 
792 static int file_to_psize(struct file *file)
793 {
794 	struct hstate *hstate = hstate_file(file);
795 	return shift_to_mmu_psize(huge_page_shift(hstate));
796 }
797 
798 unsigned long hugetlb_get_unmapped_area(struct file *file, unsigned long addr,
799 					unsigned long len, unsigned long pgoff,
800 					unsigned long flags)
801 {
802 	if (radix_enabled())
803 		return generic_hugetlb_get_unmapped_area(file, addr, len, pgoff, flags);
804 
805 	return slice_get_unmapped_area(addr, len, flags, file_to_psize(file), 1);
806 }
807 #endif
808