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_and_bitmap - get temp pages array and bitmap
24
* @chunk: chunk of interest
25
* @bitmapp: output parameter for bitmap
26
* @may_alloc: may allocate the array
28
* Returns pointer to array of pointers to struct page and bitmap,
29
* both of which can be indexed with pcpu_page_idx(). The returned
30
* array is cleared to zero and *@bitmapp is copied from
31
* @chunk->populated. Note that there is only one array and bitmap
32
* and access exclusion is the caller's responsibility.
35
* pcpu_alloc_mutex and does GFP_KERNEL allocation if @may_alloc.
36
* Otherwise, don't care.
39
* Pointer to temp pages array on success, NULL on failure.
41
static struct page **pcpu_get_pages_and_bitmap(struct pcpu_chunk *chunk,
42
unsigned long **bitmapp,
45
static struct page **pages;
46
static unsigned long *bitmap;
47
size_t pages_size = pcpu_nr_units * pcpu_unit_pages * sizeof(pages[0]);
48
size_t bitmap_size = BITS_TO_LONGS(pcpu_unit_pages) *
49
sizeof(unsigned long);
51
if (!pages || !bitmap) {
52
if (may_alloc && !pages)
53
pages = pcpu_mem_zalloc(pages_size);
54
if (may_alloc && !bitmap)
55
bitmap = pcpu_mem_zalloc(bitmap_size);
56
if (!pages || !bitmap)
60
bitmap_copy(bitmap, chunk->populated, pcpu_unit_pages);
67
* pcpu_free_pages - free pages which were allocated for @chunk
68
* @chunk: chunk pages were allocated for
69
* @pages: array of pages to be freed, indexed by pcpu_page_idx()
70
* @populated: populated bitmap
71
* @page_start: page index of the first page to be freed
72
* @page_end: page index of the last page to be freed + 1
74
* Free pages [@page_start and @page_end) in @pages for all units.
75
* The pages were allocated for @chunk.
77
static void pcpu_free_pages(struct pcpu_chunk *chunk,
78
struct page **pages, unsigned long *populated,
79
int page_start, int page_end)
84
for_each_possible_cpu(cpu) {
85
for (i = page_start; i < page_end; i++) {
86
struct page *page = pages[pcpu_page_idx(cpu, i)];
95
* pcpu_alloc_pages - allocates pages for @chunk
96
* @chunk: target chunk
97
* @pages: array to put the allocated pages into, indexed by pcpu_page_idx()
98
* @populated: populated bitmap
99
* @page_start: page index of the first page to be allocated
100
* @page_end: page index of the last page to be allocated + 1
102
* Allocate pages [@page_start,@page_end) into @pages for all units.
103
* The allocation is for @chunk. Percpu core doesn't care about the
104
* content of @pages and will pass it verbatim to pcpu_map_pages().
106
static int pcpu_alloc_pages(struct pcpu_chunk *chunk,
107
struct page **pages, unsigned long *populated,
108
int page_start, int page_end)
110
const gfp_t gfp = GFP_KERNEL | __GFP_HIGHMEM | __GFP_COLD;
114
for_each_possible_cpu(cpu) {
115
for (i = page_start; i < page_end; i++) {
116
struct page **pagep = &pages[pcpu_page_idx(cpu, i)];
118
*pagep = alloc_pages_node(cpu_to_node(cpu), gfp, 0);
120
pcpu_free_pages(chunk, pages, populated,
121
page_start, page_end);
130
* pcpu_pre_unmap_flush - flush cache prior to unmapping
131
* @chunk: chunk the regions to be flushed belongs to
132
* @page_start: page index of the first page to be flushed
133
* @page_end: page index of the last page to be flushed + 1
135
* Pages in [@page_start,@page_end) of @chunk are about to be
136
* unmapped. Flush cache. As each flushing trial can be very
137
* expensive, issue flush on the whole region at once rather than
138
* doing it for each cpu. This could be an overkill but is more
141
static void pcpu_pre_unmap_flush(struct pcpu_chunk *chunk,
142
int page_start, int page_end)
145
pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
146
pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
149
static void __pcpu_unmap_pages(unsigned long addr, int nr_pages)
151
unmap_kernel_range_noflush(addr, nr_pages << PAGE_SHIFT);
155
* pcpu_unmap_pages - unmap pages out of a pcpu_chunk
156
* @chunk: chunk of interest
157
* @pages: pages array which can be used to pass information to free
158
* @populated: populated bitmap
159
* @page_start: page index of the first page to unmap
160
* @page_end: page index of the last page to unmap + 1
162
* For each cpu, unmap pages [@page_start,@page_end) out of @chunk.
163
* Corresponding elements in @pages were cleared by the caller and can
164
* be used to carry information to pcpu_free_pages() which will be
165
* called after all unmaps are finished. The caller should call
166
* proper pre/post flush functions.
168
static void pcpu_unmap_pages(struct pcpu_chunk *chunk,
169
struct page **pages, unsigned long *populated,
170
int page_start, int page_end)
175
for_each_possible_cpu(cpu) {
176
for (i = page_start; i < page_end; i++) {
179
page = pcpu_chunk_page(chunk, cpu, i);
181
pages[pcpu_page_idx(cpu, i)] = page;
183
__pcpu_unmap_pages(pcpu_chunk_addr(chunk, cpu, page_start),
184
page_end - page_start);
187
for (i = page_start; i < page_end; i++)
188
__clear_bit(i, populated);
192
* pcpu_post_unmap_tlb_flush - flush TLB after unmapping
193
* @chunk: pcpu_chunk the regions to be flushed belong to
194
* @page_start: page index of the first page to be flushed
195
* @page_end: page index of the last page to be flushed + 1
197
* Pages [@page_start,@page_end) of @chunk have been unmapped. Flush
198
* TLB for the regions. This can be skipped if the area is to be
199
* returned to vmalloc as vmalloc will handle TLB flushing lazily.
201
* As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
202
* for the whole region.
204
static void pcpu_post_unmap_tlb_flush(struct pcpu_chunk *chunk,
205
int page_start, int page_end)
207
flush_tlb_kernel_range(
208
pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
209
pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
212
static int __pcpu_map_pages(unsigned long addr, struct page **pages,
215
return map_kernel_range_noflush(addr, nr_pages << PAGE_SHIFT,
220
* pcpu_map_pages - map pages into a pcpu_chunk
221
* @chunk: chunk of interest
222
* @pages: pages array containing pages to be mapped
223
* @populated: populated bitmap
224
* @page_start: page index of the first page to map
225
* @page_end: page index of the last page to map + 1
227
* For each cpu, map pages [@page_start,@page_end) into @chunk. The
228
* caller is responsible for calling pcpu_post_map_flush() after all
229
* mappings are complete.
231
* This function is responsible for setting corresponding bits in
232
* @chunk->populated bitmap and whatever is necessary for reverse
233
* lookup (addr -> chunk).
235
static int pcpu_map_pages(struct pcpu_chunk *chunk,
236
struct page **pages, unsigned long *populated,
237
int page_start, int page_end)
239
unsigned int cpu, tcpu;
242
for_each_possible_cpu(cpu) {
243
err = __pcpu_map_pages(pcpu_chunk_addr(chunk, cpu, page_start),
244
&pages[pcpu_page_idx(cpu, page_start)],
245
page_end - page_start);
250
/* mapping successful, link chunk and mark populated */
251
for (i = page_start; i < page_end; i++) {
252
for_each_possible_cpu(cpu)
253
pcpu_set_page_chunk(pages[pcpu_page_idx(cpu, i)],
255
__set_bit(i, populated);
261
for_each_possible_cpu(tcpu) {
264
__pcpu_unmap_pages(pcpu_chunk_addr(chunk, tcpu, page_start),
265
page_end - page_start);
271
* pcpu_post_map_flush - flush cache after mapping
272
* @chunk: pcpu_chunk the regions to be flushed belong to
273
* @page_start: page index of the first page to be flushed
274
* @page_end: page index of the last page to be flushed + 1
276
* Pages [@page_start,@page_end) of @chunk have been mapped. Flush
279
* As with pcpu_pre_unmap_flush(), TLB flushing also is done at once
280
* for the whole region.
282
static void pcpu_post_map_flush(struct pcpu_chunk *chunk,
283
int page_start, int page_end)
286
pcpu_chunk_addr(chunk, pcpu_low_unit_cpu, page_start),
287
pcpu_chunk_addr(chunk, pcpu_high_unit_cpu, page_end));
291
* pcpu_populate_chunk - populate and map an area of a pcpu_chunk
292
* @chunk: chunk of interest
293
* @off: offset to the area to populate
294
* @size: size of the area to populate in bytes
296
* For each cpu, populate and map pages [@page_start,@page_end) into
297
* @chunk. The area is cleared on return.
300
* pcpu_alloc_mutex, does GFP_KERNEL allocation.
302
static int pcpu_populate_chunk(struct pcpu_chunk *chunk, int off, int size)
304
int page_start = PFN_DOWN(off);
305
int page_end = PFN_UP(off + size);
306
int free_end = page_start, unmap_end = page_start;
308
unsigned long *populated;
312
/* quick path, check whether all pages are already there */
314
pcpu_next_pop(chunk, &rs, &re, page_end);
315
if (rs == page_start && re == page_end)
318
/* need to allocate and map pages, this chunk can't be immutable */
319
WARN_ON(chunk->immutable);
321
pages = pcpu_get_pages_and_bitmap(chunk, &populated, true);
326
pcpu_for_each_unpop_region(chunk, rs, re, page_start, page_end) {
327
rc = pcpu_alloc_pages(chunk, pages, populated, rs, re);
333
pcpu_for_each_unpop_region(chunk, rs, re, page_start, page_end) {
334
rc = pcpu_map_pages(chunk, pages, populated, rs, re);
339
pcpu_post_map_flush(chunk, page_start, page_end);
341
/* commit new bitmap */
342
bitmap_copy(chunk->populated, populated, pcpu_unit_pages);
344
for_each_possible_cpu(cpu)
345
memset((void *)pcpu_chunk_addr(chunk, cpu, 0) + off, 0, size);
349
pcpu_pre_unmap_flush(chunk, page_start, unmap_end);
350
pcpu_for_each_unpop_region(chunk, rs, re, page_start, unmap_end)
351
pcpu_unmap_pages(chunk, pages, populated, rs, re);
352
pcpu_post_unmap_tlb_flush(chunk, page_start, unmap_end);
354
pcpu_for_each_unpop_region(chunk, rs, re, page_start, free_end)
355
pcpu_free_pages(chunk, pages, populated, rs, re);
360
* pcpu_depopulate_chunk - depopulate and unmap an area of a pcpu_chunk
361
* @chunk: chunk to depopulate
362
* @off: offset to the area to depopulate
363
* @size: size of the area to depopulate in bytes
364
* @flush: whether to flush cache and tlb or not
366
* For each cpu, depopulate and unmap pages [@page_start,@page_end)
367
* from @chunk. If @flush is true, vcache is flushed before unmapping
373
static void pcpu_depopulate_chunk(struct pcpu_chunk *chunk, int off, int size)
375
int page_start = PFN_DOWN(off);
376
int page_end = PFN_UP(off + size);
378
unsigned long *populated;
381
/* quick path, check whether it's empty already */
383
pcpu_next_unpop(chunk, &rs, &re, page_end);
384
if (rs == page_start && re == page_end)
387
/* immutable chunks can't be depopulated */
388
WARN_ON(chunk->immutable);
391
* If control reaches here, there must have been at least one
392
* successful population attempt so the temp pages array must
395
pages = pcpu_get_pages_and_bitmap(chunk, &populated, false);
399
pcpu_pre_unmap_flush(chunk, page_start, page_end);
401
pcpu_for_each_pop_region(chunk, rs, re, page_start, page_end)
402
pcpu_unmap_pages(chunk, pages, populated, rs, re);
404
/* no need to flush tlb, vmalloc will handle it lazily */
406
pcpu_for_each_pop_region(chunk, rs, re, page_start, page_end)
407
pcpu_free_pages(chunk, pages, populated, rs, re);
409
/* commit new bitmap */
410
bitmap_copy(chunk->populated, populated, pcpu_unit_pages);
413
static struct pcpu_chunk *pcpu_create_chunk(void)
415
struct pcpu_chunk *chunk;
416
struct vm_struct **vms;
418
chunk = pcpu_alloc_chunk();
422
vms = pcpu_get_vm_areas(pcpu_group_offsets, pcpu_group_sizes,
423
pcpu_nr_groups, pcpu_atom_size);
425
pcpu_free_chunk(chunk);
430
chunk->base_addr = vms[0]->addr - pcpu_group_offsets[0];
434
static void pcpu_destroy_chunk(struct pcpu_chunk *chunk)
436
if (chunk && chunk->data)
437
pcpu_free_vm_areas(chunk->data, pcpu_nr_groups);
438
pcpu_free_chunk(chunk);
441
static struct page *pcpu_addr_to_page(void *addr)
443
return vmalloc_to_page(addr);
446
static int __init pcpu_verify_alloc_info(const struct pcpu_alloc_info *ai)
448
/* no extra restriction */