1/*
2 * Contiguous Memory Allocator
3 *
4 * Copyright (c) 2010-2011 by Samsung Electronics.
5 * Copyright IBM Corporation, 2013
6 * Copyright LG Electronics Inc., 2014
7 * Written by:
8 *	Marek Szyprowski <m.szyprowski@samsung.com>
9 *	Michal Nazarewicz <mina86@mina86.com>
10 *	Aneesh Kumar K.V <aneesh.kumar@linux.vnet.ibm.com>
11 *	Joonsoo Kim <iamjoonsoo.kim@lge.com>
12 *
13 * This program is free software; you can redistribute it and/or
14 * modify it under the terms of the GNU General Public License as
15 * published by the Free Software Foundation; either version 2 of the
16 * License or (at your optional) any later version of the license.
17 */
18
19#define pr_fmt(fmt) "cma: " fmt
20
21#ifdef CONFIG_CMA_DEBUG
22#ifndef DEBUG
23#  define DEBUG
24#endif
25#endif
26#define CREATE_TRACE_POINTS
27
28#include <linux/memblock.h>
29#include <linux/err.h>
30#include <linux/mm.h>
31#include <linux/mutex.h>
32#include <linux/sizes.h>
33#include <linux/slab.h>
34#include <linux/log2.h>
35#include <linux/cma.h>
36#include <linux/highmem.h>
37#include <linux/io.h>
38#include <trace/events/cma.h>
39
40#include "cma.h"
41
42struct cma cma_areas[MAX_CMA_AREAS];
43unsigned cma_area_count;
44static DEFINE_MUTEX(cma_mutex);
45
46phys_addr_t cma_get_base(const struct cma *cma)
47{
48	return PFN_PHYS(cma->base_pfn);
49}
50
51unsigned long cma_get_size(const struct cma *cma)
52{
53	return cma->count << PAGE_SHIFT;
54}
55
56static unsigned long cma_bitmap_aligned_mask(const struct cma *cma,
57					     int align_order)
58{
59	if (align_order <= cma->order_per_bit)
60		return 0;
61	return (1UL << (align_order - cma->order_per_bit)) - 1;
62}
63
64/*
65 * Find a PFN aligned to the specified order and return an offset represented in
66 * order_per_bits.
67 */
68static unsigned long cma_bitmap_aligned_offset(const struct cma *cma,
69					       int align_order)
70{
71	if (align_order <= cma->order_per_bit)
72		return 0;
73
74	return (ALIGN(cma->base_pfn, (1UL << align_order))
75		- cma->base_pfn) >> cma->order_per_bit;
76}
77
78static unsigned long cma_bitmap_pages_to_bits(const struct cma *cma,
79					      unsigned long pages)
80{
81	return ALIGN(pages, 1UL << cma->order_per_bit) >> cma->order_per_bit;
82}
83
84static void cma_clear_bitmap(struct cma *cma, unsigned long pfn,
85			     unsigned int count)
86{
87	unsigned long bitmap_no, bitmap_count;
88
89	bitmap_no = (pfn - cma->base_pfn) >> cma->order_per_bit;
90	bitmap_count = cma_bitmap_pages_to_bits(cma, count);
91
92	mutex_lock(&cma->lock);
93	bitmap_clear(cma->bitmap, bitmap_no, bitmap_count);
94	mutex_unlock(&cma->lock);
95}
96
97static int __init cma_activate_area(struct cma *cma)
98{
99	int bitmap_size = BITS_TO_LONGS(cma_bitmap_maxno(cma)) * sizeof(long);
100	unsigned long base_pfn = cma->base_pfn, pfn = base_pfn;
101	unsigned i = cma->count >> pageblock_order;
102	struct zone *zone;
103
104	cma->bitmap = kzalloc(bitmap_size, GFP_KERNEL);
105
106	if (!cma->bitmap)
107		return -ENOMEM;
108
109	WARN_ON_ONCE(!pfn_valid(pfn));
110	zone = page_zone(pfn_to_page(pfn));
111
112	do {
113		unsigned j;
114
115		base_pfn = pfn;
116		for (j = pageblock_nr_pages; j; --j, pfn++) {
117			WARN_ON_ONCE(!pfn_valid(pfn));
118			/*
119			 * alloc_contig_range requires the pfn range
120			 * specified to be in the same zone. Make this
121			 * simple by forcing the entire CMA resv range
122			 * to be in the same zone.
123			 */
124			if (page_zone(pfn_to_page(pfn)) != zone)
125				goto err;
126		}
127		init_cma_reserved_pageblock(pfn_to_page(base_pfn));
128	} while (--i);
129
130	mutex_init(&cma->lock);
131
132#ifdef CONFIG_CMA_DEBUGFS
133	INIT_HLIST_HEAD(&cma->mem_head);
134	spin_lock_init(&cma->mem_head_lock);
135#endif
136
137	return 0;
138
139err:
140	kfree(cma->bitmap);
141	cma->count = 0;
142	return -EINVAL;
143}
144
145static int __init cma_init_reserved_areas(void)
146{
147	int i;
148
149	for (i = 0; i < cma_area_count; i++) {
150		int ret = cma_activate_area(&cma_areas[i]);
151
152		if (ret)
153			return ret;
154	}
155
156	return 0;
157}
158core_initcall(cma_init_reserved_areas);
159
160/**
161 * cma_init_reserved_mem() - create custom contiguous area from reserved memory
162 * @base: Base address of the reserved area
163 * @size: Size of the reserved area (in bytes),
164 * @order_per_bit: Order of pages represented by one bit on bitmap.
165 * @res_cma: Pointer to store the created cma region.
166 *
167 * This function creates custom contiguous area from already reserved memory.
168 */
169int __init cma_init_reserved_mem(phys_addr_t base, phys_addr_t size,
170				 unsigned int order_per_bit,
171				 struct cma **res_cma)
172{
173	struct cma *cma;
174	phys_addr_t alignment;
175
176	/* Sanity checks */
177	if (cma_area_count == ARRAY_SIZE(cma_areas)) {
178		pr_err("Not enough slots for CMA reserved regions!\n");
179		return -ENOSPC;
180	}
181
182	if (!size || !memblock_is_region_reserved(base, size))
183		return -EINVAL;
184
185	/* ensure minimal alignment requied by mm core */
186	alignment = PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order);
187
188	/* alignment should be aligned with order_per_bit */
189	if (!IS_ALIGNED(alignment >> PAGE_SHIFT, 1 << order_per_bit))
190		return -EINVAL;
191
192	if (ALIGN(base, alignment) != base || ALIGN(size, alignment) != size)
193		return -EINVAL;
194
195	/*
196	 * Each reserved area must be initialised later, when more kernel
197	 * subsystems (like slab allocator) are available.
198	 */
199	cma = &cma_areas[cma_area_count];
200	cma->base_pfn = PFN_DOWN(base);
201	cma->count = size >> PAGE_SHIFT;
202	cma->order_per_bit = order_per_bit;
203	*res_cma = cma;
204	cma_area_count++;
205	totalcma_pages += (size / PAGE_SIZE);
206
207	return 0;
208}
209
210/**
211 * cma_declare_contiguous() - reserve custom contiguous area
212 * @base: Base address of the reserved area optional, use 0 for any
213 * @size: Size of the reserved area (in bytes),
214 * @limit: End address of the reserved memory (optional, 0 for any).
215 * @alignment: Alignment for the CMA area, should be power of 2 or zero
216 * @order_per_bit: Order of pages represented by one bit on bitmap.
217 * @fixed: hint about where to place the reserved area
218 * @res_cma: Pointer to store the created cma region.
219 *
220 * This function reserves memory from early allocator. It should be
221 * called by arch specific code once the early allocator (memblock or bootmem)
222 * has been activated and all other subsystems have already allocated/reserved
223 * memory. This function allows to create custom reserved areas.
224 *
225 * If @fixed is true, reserve contiguous area at exactly @base.  If false,
226 * reserve in range from @base to @limit.
227 */
228int __init cma_declare_contiguous(phys_addr_t base,
229			phys_addr_t size, phys_addr_t limit,
230			phys_addr_t alignment, unsigned int order_per_bit,
231			bool fixed, struct cma **res_cma)
232{
233	phys_addr_t memblock_end = memblock_end_of_DRAM();
234	phys_addr_t highmem_start;
235	int ret = 0;
236
237#ifdef CONFIG_X86
238	/*
239	 * high_memory isn't direct mapped memory so retrieving its physical
240	 * address isn't appropriate.  But it would be useful to check the
241	 * physical address of the highmem boundary so it's justfiable to get
242	 * the physical address from it.  On x86 there is a validation check for
243	 * this case, so the following workaround is needed to avoid it.
244	 */
245	highmem_start = __pa_nodebug(high_memory);
246#else
247	highmem_start = __pa(high_memory);
248#endif
249	pr_debug("%s(size %pa, base %pa, limit %pa alignment %pa)\n",
250		__func__, &size, &base, &limit, &alignment);
251
252	if (cma_area_count == ARRAY_SIZE(cma_areas)) {
253		pr_err("Not enough slots for CMA reserved regions!\n");
254		return -ENOSPC;
255	}
256
257	if (!size)
258		return -EINVAL;
259
260	if (alignment && !is_power_of_2(alignment))
261		return -EINVAL;
262
263	/*
264	 * Sanitise input arguments.
265	 * Pages both ends in CMA area could be merged into adjacent unmovable
266	 * migratetype page by page allocator's buddy algorithm. In the case,
267	 * you couldn't get a contiguous memory, which is not what we want.
268	 */
269	alignment = max(alignment,
270		(phys_addr_t)PAGE_SIZE << max(MAX_ORDER - 1, pageblock_order));
271	base = ALIGN(base, alignment);
272	size = ALIGN(size, alignment);
273	limit &= ~(alignment - 1);
274
275	if (!base)
276		fixed = false;
277
278	/* size should be aligned with order_per_bit */
279	if (!IS_ALIGNED(size >> PAGE_SHIFT, 1 << order_per_bit))
280		return -EINVAL;
281
282	/*
283	 * If allocating at a fixed base the request region must not cross the
284	 * low/high memory boundary.
285	 */
286	if (fixed && base < highmem_start && base + size > highmem_start) {
287		ret = -EINVAL;
288		pr_err("Region at %pa defined on low/high memory boundary (%pa)\n",
289			&base, &highmem_start);
290		goto err;
291	}
292
293	/*
294	 * If the limit is unspecified or above the memblock end, its effective
295	 * value will be the memblock end. Set it explicitly to simplify further
296	 * checks.
297	 */
298	if (limit == 0 || limit > memblock_end)
299		limit = memblock_end;
300
301	/* Reserve memory */
302	if (fixed) {
303		if (memblock_is_region_reserved(base, size) ||
304		    memblock_reserve(base, size) < 0) {
305			ret = -EBUSY;
306			goto err;
307		}
308	} else {
309		phys_addr_t addr = 0;
310
311		/*
312		 * All pages in the reserved area must come from the same zone.
313		 * If the requested region crosses the low/high memory boundary,
314		 * try allocating from high memory first and fall back to low
315		 * memory in case of failure.
316		 */
317		if (base < highmem_start && limit > highmem_start) {
318			addr = memblock_alloc_range(size, alignment,
319						    highmem_start, limit);
320			limit = highmem_start;
321		}
322
323		if (!addr) {
324			addr = memblock_alloc_range(size, alignment, base,
325						    limit);
326			if (!addr) {
327				ret = -ENOMEM;
328				goto err;
329			}
330		}
331
332		/*
333		 * kmemleak scans/reads tracked objects for pointers to other
334		 * objects but this address isn't mapped and accessible
335		 */
336		kmemleak_ignore(phys_to_virt(addr));
337		base = addr;
338	}
339
340	ret = cma_init_reserved_mem(base, size, order_per_bit, res_cma);
341	if (ret)
342		goto err;
343
344	pr_info("Reserved %ld MiB at %pa\n", (unsigned long)size / SZ_1M,
345		&base);
346	return 0;
347
348err:
349	pr_err("Failed to reserve %ld MiB\n", (unsigned long)size / SZ_1M);
350	return ret;
351}
352
353/**
354 * cma_alloc() - allocate pages from contiguous area
355 * @cma:   Contiguous memory region for which the allocation is performed.
356 * @count: Requested number of pages.
357 * @align: Requested alignment of pages (in PAGE_SIZE order).
358 *
359 * This function allocates part of contiguous memory on specific
360 * contiguous memory area.
361 */
362struct page *cma_alloc(struct cma *cma, unsigned int count, unsigned int align)
363{
364	unsigned long mask, offset, pfn, start = 0;
365	unsigned long bitmap_maxno, bitmap_no, bitmap_count;
366	struct page *page = NULL;
367	int ret;
368
369	if (!cma || !cma->count)
370		return NULL;
371
372	pr_debug("%s(cma %p, count %d, align %d)\n", __func__, (void *)cma,
373		 count, align);
374
375	if (!count)
376		return NULL;
377
378	mask = cma_bitmap_aligned_mask(cma, align);
379	offset = cma_bitmap_aligned_offset(cma, align);
380	bitmap_maxno = cma_bitmap_maxno(cma);
381	bitmap_count = cma_bitmap_pages_to_bits(cma, count);
382
383	for (;;) {
384		mutex_lock(&cma->lock);
385		bitmap_no = bitmap_find_next_zero_area_off(cma->bitmap,
386				bitmap_maxno, start, bitmap_count, mask,
387				offset);
388		if (bitmap_no >= bitmap_maxno) {
389			mutex_unlock(&cma->lock);
390			break;
391		}
392		bitmap_set(cma->bitmap, bitmap_no, bitmap_count);
393		/*
394		 * It's safe to drop the lock here. We've marked this region for
395		 * our exclusive use. If the migration fails we will take the
396		 * lock again and unmark it.
397		 */
398		mutex_unlock(&cma->lock);
399
400		pfn = cma->base_pfn + (bitmap_no << cma->order_per_bit);
401		mutex_lock(&cma_mutex);
402		ret = alloc_contig_range(pfn, pfn + count, MIGRATE_CMA);
403		mutex_unlock(&cma_mutex);
404		if (ret == 0) {
405			page = pfn_to_page(pfn);
406			break;
407		}
408
409		cma_clear_bitmap(cma, pfn, count);
410		if (ret != -EBUSY)
411			break;
412
413		pr_debug("%s(): memory range at %p is busy, retrying\n",
414			 __func__, pfn_to_page(pfn));
415		/* try again with a bit different memory target */
416		start = bitmap_no + mask + 1;
417	}
418
419	trace_cma_alloc(page ? pfn : -1UL, page, count, align);
420
421	pr_debug("%s(): returned %p\n", __func__, page);
422	return page;
423}
424
425/**
426 * cma_release() - release allocated pages
427 * @cma:   Contiguous memory region for which the allocation is performed.
428 * @pages: Allocated pages.
429 * @count: Number of allocated pages.
430 *
431 * This function releases memory allocated by alloc_cma().
432 * It returns false when provided pages do not belong to contiguous area and
433 * true otherwise.
434 */
435bool cma_release(struct cma *cma, const struct page *pages, unsigned int count)
436{
437	unsigned long pfn;
438
439	if (!cma || !pages)
440		return false;
441
442	pr_debug("%s(page %p)\n", __func__, (void *)pages);
443
444	pfn = page_to_pfn(pages);
445
446	if (pfn < cma->base_pfn || pfn >= cma->base_pfn + cma->count)
447		return false;
448
449	VM_BUG_ON(pfn + count > cma->base_pfn + cma->count);
450
451	free_contig_range(pfn, count);
452	cma_clear_bitmap(cma, pfn, count);
453	trace_cma_release(pfn, pages, count);
454
455	return true;
456}
457