1 // SPDX-License-Identifier: GPL-2.0-only
3 * mm/percpu-vm.c - vmalloc area based chunk allocation
5 * Copyright (C) 2010 SUSE Linux Products GmbH
6 * Copyright (C) 2010 Tejun Heo <tj@kernel.org>
8 * Chunks are mapped into vmalloc areas and populated page by page.
9 * This is the default chunk allocator.
12 static struct page *pcpu_chunk_page(struct pcpu_chunk *chunk,
13 unsigned int cpu, int page_idx)
15 /* must not be used on pre-mapped chunk */
16 WARN_ON(chunk->immutable);
18 return vmalloc_to_page((void *)pcpu_chunk_addr(chunk, cpu, page_idx));
22 * pcpu_get_pages - get temp pages array
24 * Returns pointer to array of pointers to struct page which can be indexed
25 * with pcpu_page_idx(). Note that there is only one array and accesses
26 * should be serialized by pcpu_alloc_mutex.
29 * Pointer to temp pages array on success.
31 static struct page **pcpu_get_pages(void)
33 static struct page **pages;
34 size_t pages_size = pcpu_nr_units * pcpu_unit_pages * sizeof(pages[0]);
36 lockdep_assert_held(&pcpu_alloc_mutex);
39 pages = pcpu_mem_zalloc(pages_size, GFP_KERNEL);
44 * pcpu_free_pages - free pages which were allocated for @chunk
45 * @chunk: chunk pages were allocated for
46 * @pages: array of pages to be freed, indexed by pcpu_page_idx()
47 * @page_start: page index of the first page to be freed
48 * @page_end: page index of the last page to be freed + 1
50 * Free pages [@page_start and @page_end) in @pages for all units.
51 * The pages were allocated for @chunk.
53 static void pcpu_free_pages(struct pcpu_chunk *chunk,
54 struct page **pages, int page_start, int page_end)
59 for_each_possible_cpu(cpu) {
60 for (i = page_start; i < page_end; i++) {
61 struct page *page = pages[pcpu_page_idx(cpu, i)];
70 * pcpu_alloc_pages - allocates pages for @chunk
71 * @chunk: target chunk
72 * @pages: array to put the allocated pages into, indexed by pcpu_page_idx()
73 * @page_start: page index of the first page to be allocated
74 * @page_end: page index of the last page to be allocated + 1
75 * @gfp: allocation flags passed to the underlying allocator
77 * Allocate pages [@page_start,@page_end) into @pages for all units.
78 * The allocation is for @chunk. Percpu core doesn't care about the
79 * content of @pages and will pass it verbatim to pcpu_map_pages().
81 static int pcpu_alloc_pages(struct pcpu_chunk *chunk,
82 struct page **pages, int page_start, int page_end,
85 unsigned int cpu, tcpu;
90 for_each_possible_cpu(cpu) {
91 for (i = page_start; i < page_end; i++) {
92 struct page **pagep = &pages[pcpu_page_idx(cpu, i)];
94 *pagep = alloc_pages_node(cpu_to_node(cpu), gfp, 0);
102 while (--i >= page_start)
103 __free_page(pages[pcpu_page_idx(cpu, i)]);
105 for_each_possible_cpu(tcpu) {
108 for (i = page_start; i < page_end; i++)
109 __free_page(pages[pcpu_page_idx(tcpu, i)]);
115 * pcpu_pre_unmap_flush - flush cache prior to unmapping
116 * @chunk: chunk the regions to be flushed belongs to
117 * @page_start: page index of the first page to be flushed
118 * @page_end: page index of the last page to be flushed + 1
120 * Pages in [@page_start,@page_end) of @chunk are about to be
121 * unmapped. Flush cache. As each flushing trial can be very
122 * expensive, issue flush on the whole region at once rather than
123 * doing it for each cpu. This could be an overkill but is more
126 static void pcpu_pre_unmap_flush(struct pcpu_chunk *chunk,
127 int page_start, int page_end)
130 pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
131 pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
134 static void __pcpu_unmap_pages(unsigned long addr, int nr_pages)
136 unmap_kernel_range_noflush(addr, nr_pages << PAGE_SHIFT);
140 * pcpu_unmap_pages - unmap pages out of a pcpu_chunk
141 * @chunk: chunk of interest
142 * @pages: pages array which can be used to pass information to free
143 * @page_start: page index of the first page to unmap
144 * @page_end: page index of the last page to unmap + 1
146 * For each cpu, unmap pages [@page_start,@page_end) out of @chunk.
147 * Corresponding elements in @pages were cleared by the caller and can
148 * be used to carry information to pcpu_free_pages() which will be
149 * called after all unmaps are finished. The caller should call
150 * proper pre/post flush functions.
152 static void pcpu_unmap_pages(struct pcpu_chunk *chunk,
153 struct page **pages, int page_start, int page_end)
158 for_each_possible_cpu(cpu) {
159 for (i = page_start; i < page_end; i++) {
162 page = pcpu_chunk_page(chunk, cpu, i);
164 pages[pcpu_page_idx(cpu, i)] = page;
166 __pcpu_unmap_pages(pcpu_chunk_addr(chunk, cpu, page_start),
167 page_end - page_start);
172 * pcpu_post_unmap_tlb_flush - flush TLB after unmapping
173 * @chunk: pcpu_chunk the regions to be flushed belong to
174 * @page_start: page index of the first page to be flushed
175 * @page_end: page index of the last page to be flushed + 1
177 * Pages [@page_start,@page_end) of @chunk have been unmapped. Flush
178 * TLB for the regions. This can be skipped if the area is to be
179 * returned to vmalloc as vmalloc will handle TLB flushing lazily.
181 * As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
182 * for the whole region.
184 static void pcpu_post_unmap_tlb_flush(struct pcpu_chunk *chunk,
185 int page_start, int page_end)
187 flush_tlb_kernel_range(
188 pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
189 pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
192 static int __pcpu_map_pages(unsigned long addr, struct page **pages,
195 return map_kernel_range_noflush(addr, nr_pages << PAGE_SHIFT,
200 * pcpu_map_pages - map pages into a pcpu_chunk
201 * @chunk: chunk of interest
202 * @pages: pages array containing pages to be mapped
203 * @page_start: page index of the first page to map
204 * @page_end: page index of the last page to map + 1
206 * For each cpu, map pages [@page_start,@page_end) into @chunk. The
207 * caller is responsible for calling pcpu_post_map_flush() after all
208 * mappings are complete.
210 * This function is responsible for setting up whatever is necessary for
211 * reverse lookup (addr -> chunk).
213 static int pcpu_map_pages(struct pcpu_chunk *chunk,
214 struct page **pages, int page_start, int page_end)
216 unsigned int cpu, tcpu;
219 for_each_possible_cpu(cpu) {
220 err = __pcpu_map_pages(pcpu_chunk_addr(chunk, cpu, page_start),
221 &pages[pcpu_page_idx(cpu, page_start)],
222 page_end - page_start);
226 for (i = page_start; i < page_end; i++)
227 pcpu_set_page_chunk(pages[pcpu_page_idx(cpu, i)],
232 for_each_possible_cpu(tcpu) {
235 __pcpu_unmap_pages(pcpu_chunk_addr(chunk, tcpu, page_start),
236 page_end - page_start);
238 pcpu_post_unmap_tlb_flush(chunk, page_start, page_end);
243 * pcpu_post_map_flush - flush cache after mapping
244 * @chunk: pcpu_chunk the regions to be flushed belong to
245 * @page_start: page index of the first page to be flushed
246 * @page_end: page index of the last page to be flushed + 1
248 * Pages [@page_start,@page_end) of @chunk have been mapped. Flush
251 * As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
252 * for the whole region.
254 static void pcpu_post_map_flush(struct pcpu_chunk *chunk,
255 int page_start, int page_end)
258 pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
259 pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
263 * pcpu_populate_chunk - populate and map an area of a pcpu_chunk
264 * @chunk: chunk of interest
265 * @page_start: the start page
266 * @page_end: the end page
267 * @gfp: allocation flags passed to the underlying memory allocator
269 * For each cpu, populate and map pages [@page_start,@page_end) into
273 * pcpu_alloc_mutex, does GFP_KERNEL allocation.
275 static int pcpu_populate_chunk(struct pcpu_chunk *chunk,
276 int page_start, int page_end, gfp_t gfp)
280 pages = pcpu_get_pages();
284 if (pcpu_alloc_pages(chunk, pages, page_start, page_end, gfp))
287 if (pcpu_map_pages(chunk, pages, page_start, page_end)) {
288 pcpu_free_pages(chunk, pages, page_start, page_end);
291 pcpu_post_map_flush(chunk, page_start, page_end);
297 * pcpu_depopulate_chunk - depopulate and unmap an area of a pcpu_chunk
298 * @chunk: chunk to depopulate
299 * @page_start: the start page
300 * @page_end: the end page
302 * For each cpu, depopulate and unmap pages [@page_start,@page_end)
308 static void pcpu_depopulate_chunk(struct pcpu_chunk *chunk,
309 int page_start, int page_end)
314 * If control reaches here, there must have been at least one
315 * successful population attempt so the temp pages array must
318 pages = pcpu_get_pages();
322 pcpu_pre_unmap_flush(chunk, page_start, page_end);
324 pcpu_unmap_pages(chunk, pages, page_start, page_end);
326 /* no need to flush tlb, vmalloc will handle it lazily */
328 pcpu_free_pages(chunk, pages, page_start, page_end);
331 static struct pcpu_chunk *pcpu_create_chunk(enum pcpu_chunk_type type,
334 struct pcpu_chunk *chunk;
335 struct vm_struct **vms;
337 chunk = pcpu_alloc_chunk(type, gfp);
341 vms = pcpu_get_vm_areas(pcpu_group_offsets, pcpu_group_sizes,
342 pcpu_nr_groups, pcpu_atom_size);
344 pcpu_free_chunk(chunk);
349 chunk->base_addr = vms[0]->addr - pcpu_group_offsets[0];
351 pcpu_stats_chunk_alloc();
352 trace_percpu_create_chunk(chunk->base_addr);
357 static void pcpu_destroy_chunk(struct pcpu_chunk *chunk)
362 pcpu_stats_chunk_dealloc();
363 trace_percpu_destroy_chunk(chunk->base_addr);
366 pcpu_free_vm_areas(chunk->data, pcpu_nr_groups);
367 pcpu_free_chunk(chunk);
370 static struct page *pcpu_addr_to_page(void *addr)
372 return vmalloc_to_page(addr);
375 static int __init pcpu_verify_alloc_info(const struct pcpu_alloc_info *ai)
377 /* no extra restriction */