2 * mm/percpu-vm.c - vmalloc area based chunk allocation
4 * Copyright (C) 2010 SUSE Linux Products GmbH
5 * Copyright (C) 2010 Tejun Heo <tj@kernel.org>
7 * This file is released under the GPLv2.
9 * Chunks are mapped into vmalloc areas and populated page by page.
10 * This is the default chunk allocator.
13 static struct page *pcpu_chunk_page(struct pcpu_chunk *chunk,
14 unsigned int cpu, int page_idx)
16 /* must not be used on pre-mapped chunk */
17 WARN_ON(chunk->immutable);
19 return vmalloc_to_page((void *)pcpu_chunk_addr(chunk, cpu, page_idx));
23 * pcpu_get_pages - get temp pages array
24 * @chunk: chunk of interest
26 * Returns pointer to array of pointers to struct page which can be indexed
27 * with pcpu_page_idx(). Note that there is only one array and accesses
28 * should be serialized by pcpu_alloc_mutex.
31 * Pointer to temp pages array on success.
33 static struct page **pcpu_get_pages(struct pcpu_chunk *chunk_alloc)
35 static struct page **pages;
36 size_t pages_size = pcpu_nr_units * pcpu_unit_pages * sizeof(pages[0]);
38 lockdep_assert_held(&pcpu_alloc_mutex);
41 pages = pcpu_mem_zalloc(pages_size);
46 * pcpu_free_pages - free pages which were allocated for @chunk
47 * @chunk: chunk pages were allocated for
48 * @pages: array of pages to be freed, indexed by pcpu_page_idx()
49 * @page_start: page index of the first page to be freed
50 * @page_end: page index of the last page to be freed + 1
52 * Free pages [@page_start and @page_end) in @pages for all units.
53 * The pages were allocated for @chunk.
55 static void pcpu_free_pages(struct pcpu_chunk *chunk,
56 struct page **pages, int page_start, int page_end)
61 for_each_possible_cpu(cpu) {
62 for (i = page_start; i < page_end; i++) {
63 struct page *page = pages[pcpu_page_idx(cpu, i)];
72 * pcpu_alloc_pages - allocates pages for @chunk
73 * @chunk: target chunk
74 * @pages: array to put the allocated pages into, indexed by pcpu_page_idx()
75 * @page_start: page index of the first page to be allocated
76 * @page_end: page index of the last page to be allocated + 1
78 * Allocate pages [@page_start,@page_end) into @pages for all units.
79 * The allocation is for @chunk. Percpu core doesn't care about the
80 * content of @pages and will pass it verbatim to pcpu_map_pages().
82 static int pcpu_alloc_pages(struct pcpu_chunk *chunk,
83 struct page **pages, int page_start, int page_end)
85 const gfp_t gfp = GFP_KERNEL | __GFP_HIGHMEM | __GFP_COLD;
86 unsigned int cpu, tcpu;
89 for_each_possible_cpu(cpu) {
90 for (i = page_start; i < page_end; i++) {
91 struct page **pagep = &pages[pcpu_page_idx(cpu, i)];
93 *pagep = alloc_pages_node(cpu_to_node(cpu), gfp, 0);
101 while (--i >= page_start)
102 __free_page(pages[pcpu_page_idx(cpu, i)]);
104 for_each_possible_cpu(tcpu) {
107 for (i = page_start; i < page_end; i++)
108 __free_page(pages[pcpu_page_idx(tcpu, i)]);
114 * pcpu_pre_unmap_flush - flush cache prior to unmapping
115 * @chunk: chunk the regions to be flushed belongs to
116 * @page_start: page index of the first page to be flushed
117 * @page_end: page index of the last page to be flushed + 1
119 * Pages in [@page_start,@page_end) of @chunk are about to be
120 * unmapped. Flush cache. As each flushing trial can be very
121 * expensive, issue flush on the whole region at once rather than
122 * doing it for each cpu. This could be an overkill but is more
125 static void pcpu_pre_unmap_flush(struct pcpu_chunk *chunk,
126 int page_start, int page_end)
129 pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
130 pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
133 static void __pcpu_unmap_pages(unsigned long addr, int nr_pages)
135 unmap_kernel_range_noflush(addr, nr_pages << PAGE_SHIFT);
139 * pcpu_unmap_pages - unmap pages out of a pcpu_chunk
140 * @chunk: chunk of interest
141 * @pages: pages array which can be used to pass information to free
142 * @page_start: page index of the first page to unmap
143 * @page_end: page index of the last page to unmap + 1
145 * For each cpu, unmap pages [@page_start,@page_end) out of @chunk.
146 * Corresponding elements in @pages were cleared by the caller and can
147 * be used to carry information to pcpu_free_pages() which will be
148 * called after all unmaps are finished. The caller should call
149 * proper pre/post flush functions.
151 static void pcpu_unmap_pages(struct pcpu_chunk *chunk,
152 struct page **pages, int page_start, int page_end)
157 for_each_possible_cpu(cpu) {
158 for (i = page_start; i < page_end; i++) {
161 page = pcpu_chunk_page(chunk, cpu, i);
163 pages[pcpu_page_idx(cpu, i)] = page;
165 __pcpu_unmap_pages(pcpu_chunk_addr(chunk, cpu, page_start),
166 page_end - page_start);
171 * pcpu_post_unmap_tlb_flush - flush TLB after unmapping
172 * @chunk: pcpu_chunk the regions to be flushed belong to
173 * @page_start: page index of the first page to be flushed
174 * @page_end: page index of the last page to be flushed + 1
176 * Pages [@page_start,@page_end) of @chunk have been unmapped. Flush
177 * TLB for the regions. This can be skipped if the area is to be
178 * returned to vmalloc as vmalloc will handle TLB flushing lazily.
180 * As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
181 * for the whole region.
183 static void pcpu_post_unmap_tlb_flush(struct pcpu_chunk *chunk,
184 int page_start, int page_end)
186 flush_tlb_kernel_range(
187 pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
188 pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
191 static int __pcpu_map_pages(unsigned long addr, struct page **pages,
194 return map_kernel_range_noflush(addr, nr_pages << PAGE_SHIFT,
199 * pcpu_map_pages - map pages into a pcpu_chunk
200 * @chunk: chunk of interest
201 * @pages: pages array containing pages to be mapped
202 * @page_start: page index of the first page to map
203 * @page_end: page index of the last page to map + 1
205 * For each cpu, map pages [@page_start,@page_end) into @chunk. The
206 * caller is responsible for calling pcpu_post_map_flush() after all
207 * mappings are complete.
209 * This function is responsible for setting up whatever is necessary for
210 * reverse lookup (addr -> chunk).
212 static int pcpu_map_pages(struct pcpu_chunk *chunk,
213 struct page **pages, int page_start, int page_end)
215 unsigned int cpu, tcpu;
218 for_each_possible_cpu(cpu) {
219 err = __pcpu_map_pages(pcpu_chunk_addr(chunk, cpu, page_start),
220 &pages[pcpu_page_idx(cpu, page_start)],
221 page_end - page_start);
225 for (i = page_start; i < page_end; i++)
226 pcpu_set_page_chunk(pages[pcpu_page_idx(cpu, i)],
231 for_each_possible_cpu(tcpu) {
234 __pcpu_unmap_pages(pcpu_chunk_addr(chunk, tcpu, page_start),
235 page_end - page_start);
237 pcpu_post_unmap_tlb_flush(chunk, page_start, page_end);
242 * pcpu_post_map_flush - flush cache after mapping
243 * @chunk: pcpu_chunk the regions to be flushed belong to
244 * @page_start: page index of the first page to be flushed
245 * @page_end: page index of the last page to be flushed + 1
247 * Pages [@page_start,@page_end) of @chunk have been mapped. Flush
250 * As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
251 * for the whole region.
253 static void pcpu_post_map_flush(struct pcpu_chunk *chunk,
254 int page_start, int page_end)
257 pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
258 pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
262 * pcpu_populate_chunk - populate and map an area of a pcpu_chunk
263 * @chunk: chunk of interest
264 * @off: offset to the area to populate
265 * @size: size of the area to populate in bytes
267 * For each cpu, populate and map pages [@page_start,@page_end) into
268 * @chunk. The area is cleared on return.
271 * pcpu_alloc_mutex, does GFP_KERNEL allocation.
273 static int pcpu_populate_chunk(struct pcpu_chunk *chunk, int off, int size)
275 int page_start = PFN_DOWN(off);
276 int page_end = PFN_UP(off + size);
277 int free_end = page_start, unmap_end = page_start;
282 /* quick path, check whether all pages are already there */
284 pcpu_next_pop(chunk, &rs, &re, page_end);
285 if (rs == page_start && re == page_end)
288 /* need to allocate and map pages, this chunk can't be immutable */
289 WARN_ON(chunk->immutable);
291 pages = pcpu_get_pages(chunk);
296 pcpu_for_each_unpop_region(chunk, rs, re, page_start, page_end) {
297 rc = pcpu_alloc_pages(chunk, pages, rs, re);
303 pcpu_for_each_unpop_region(chunk, rs, re, page_start, page_end) {
304 rc = pcpu_map_pages(chunk, pages, rs, re);
309 pcpu_post_map_flush(chunk, page_start, page_end);
311 bitmap_set(chunk->populated, page_start, page_end - page_start);
313 for_each_possible_cpu(cpu)
314 memset((void *)pcpu_chunk_addr(chunk, cpu, 0) + off, 0, size);
318 pcpu_pre_unmap_flush(chunk, page_start, unmap_end);
319 pcpu_for_each_unpop_region(chunk, rs, re, page_start, unmap_end)
320 pcpu_unmap_pages(chunk, pages, rs, re);
321 pcpu_post_unmap_tlb_flush(chunk, page_start, unmap_end);
323 pcpu_for_each_unpop_region(chunk, rs, re, page_start, free_end)
324 pcpu_free_pages(chunk, pages, rs, re);
329 * pcpu_depopulate_chunk - depopulate and unmap an area of a pcpu_chunk
330 * @chunk: chunk to depopulate
331 * @off: offset to the area to depopulate
332 * @size: size of the area to depopulate in bytes
334 * For each cpu, depopulate and unmap pages [@page_start,@page_end)
335 * from @chunk. If @flush is true, vcache is flushed before unmapping
341 static void pcpu_depopulate_chunk(struct pcpu_chunk *chunk, int off, int size)
343 int page_start = PFN_DOWN(off);
344 int page_end = PFN_UP(off + size);
348 /* quick path, check whether it's empty already */
350 pcpu_next_unpop(chunk, &rs, &re, page_end);
351 if (rs == page_start && re == page_end)
354 /* immutable chunks can't be depopulated */
355 WARN_ON(chunk->immutable);
358 * If control reaches here, there must have been at least one
359 * successful population attempt so the temp pages array must
362 pages = pcpu_get_pages(chunk);
366 pcpu_pre_unmap_flush(chunk, page_start, page_end);
368 pcpu_for_each_pop_region(chunk, rs, re, page_start, page_end)
369 pcpu_unmap_pages(chunk, pages, rs, re);
371 /* no need to flush tlb, vmalloc will handle it lazily */
373 pcpu_for_each_pop_region(chunk, rs, re, page_start, page_end)
374 pcpu_free_pages(chunk, pages, rs, re);
376 bitmap_clear(chunk->populated, page_start, page_end - page_start);
379 static struct pcpu_chunk *pcpu_create_chunk(void)
381 struct pcpu_chunk *chunk;
382 struct vm_struct **vms;
384 chunk = pcpu_alloc_chunk();
388 vms = pcpu_get_vm_areas(pcpu_group_offsets, pcpu_group_sizes,
389 pcpu_nr_groups, pcpu_atom_size);
391 pcpu_free_chunk(chunk);
396 chunk->base_addr = vms[0]->addr - pcpu_group_offsets[0];
400 static void pcpu_destroy_chunk(struct pcpu_chunk *chunk)
402 if (chunk && chunk->data)
403 pcpu_free_vm_areas(chunk->data, pcpu_nr_groups);
404 pcpu_free_chunk(chunk);
407 static struct page *pcpu_addr_to_page(void *addr)
409 return vmalloc_to_page(addr);
412 static int __init pcpu_verify_alloc_info(const struct pcpu_alloc_info *ai)
414 /* no extra restriction */