projects/linux·blob v3.13 · refs

417 lines · 14.8 KiB

1#ifndef __LINUX_GFP_H
2#define __LINUX_GFP_H
3
4#include <linux/mmzone.h>
5#include <linux/stddef.h>
6#include <linux/linkage.h>
7#include <linux/topology.h>
8#include <linux/mmdebug.h>
9
10struct vm_area_struct;
11
12/* Plain integer GFP bitmasks. Do not use this directly. */
13#define ___GFP_DMA 0x01u
14#define ___GFP_HIGHMEM 0x02u
15#define ___GFP_DMA32 0x04u
16#define ___GFP_MOVABLE 0x08u
17#define ___GFP_WAIT 0x10u
18#define ___GFP_HIGH 0x20u
19#define ___GFP_IO 0x40u
20#define ___GFP_FS 0x80u
21#define ___GFP_COLD 0x100u
22#define ___GFP_NOWARN 0x200u
23#define ___GFP_REPEAT 0x400u
24#define ___GFP_NOFAIL 0x800u
25#define ___GFP_NORETRY 0x1000u
26#define ___GFP_MEMALLOC 0x2000u
27#define ___GFP_COMP 0x4000u
28#define ___GFP_ZERO 0x8000u
29#define ___GFP_NOMEMALLOC 0x10000u
30#define ___GFP_HARDWALL 0x20000u
31#define ___GFP_THISNODE 0x40000u
32#define ___GFP_RECLAIMABLE 0x80000u
33#define ___GFP_KMEMCG 0x100000u
34#define ___GFP_NOTRACK 0x200000u
35#define ___GFP_NO_KSWAPD 0x400000u
36#define ___GFP_OTHER_NODE 0x800000u
37#define ___GFP_WRITE 0x1000000u
38/* If the above are modified, __GFP_BITS_SHIFT may need updating */
39
40/*
41 * GFP bitmasks..
42 *
43 * Zone modifiers (see linux/mmzone.h - low three bits)
44 *
45 * Do not put any conditional on these. If necessary modify the definitions
46 * without the underscores and use them consistently. The definitions here may
47 * be used in bit comparisons.
48 */
49#define __GFP_DMA ((__force gfp_t)___GFP_DMA)
50#define __GFP_HIGHMEM ((__force gfp_t)___GFP_HIGHMEM)
51#define __GFP_DMA32 ((__force gfp_t)___GFP_DMA32)
52#define __GFP_MOVABLE ((__force gfp_t)___GFP_MOVABLE) /* Page is movable */
53#define GFP_ZONEMASK (__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
54/*
55 * Action modifiers - doesn't change the zoning
56 *
57 * __GFP_REPEAT: Try hard to allocate the memory, but the allocation attempt
58 * _might_ fail. This depends upon the particular VM implementation.
59 *
60 * __GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
61 * cannot handle allocation failures. This modifier is deprecated and no new
62 * users should be added.
63 *
64 * __GFP_NORETRY: The VM implementation must not retry indefinitely.
65 *
66 * __GFP_MOVABLE: Flag that this page will be movable by the page migration
67 * mechanism or reclaimed
68 */
69#define __GFP_WAIT ((__force gfp_t)___GFP_WAIT) /* Can wait and reschedule? */
70#define __GFP_HIGH ((__force gfp_t)___GFP_HIGH) /* Should access emergency pools? */
71#define __GFP_IO ((__force gfp_t)___GFP_IO) /* Can start physical IO? */
72#define __GFP_FS ((__force gfp_t)___GFP_FS) /* Can call down to low-level FS? */
73#define __GFP_COLD ((__force gfp_t)___GFP_COLD) /* Cache-cold page required */
74#define __GFP_NOWARN ((__force gfp_t)___GFP_NOWARN) /* Suppress page allocation failure warning */
75#define __GFP_REPEAT ((__force gfp_t)___GFP_REPEAT) /* See above */
76#define __GFP_NOFAIL ((__force gfp_t)___GFP_NOFAIL) /* See above */
77#define __GFP_NORETRY ((__force gfp_t)___GFP_NORETRY) /* See above */
78#define __GFP_MEMALLOC ((__force gfp_t)___GFP_MEMALLOC)/* Allow access to emergency reserves */
79#define __GFP_COMP ((__force gfp_t)___GFP_COMP) /* Add compound page metadata */
80#define __GFP_ZERO ((__force gfp_t)___GFP_ZERO) /* Return zeroed page on success */
81#define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC) /* Don't use emergency reserves.
82 * This takes precedence over the
83 * __GFP_MEMALLOC flag if both are
84 * set
85 */
86#define __GFP_HARDWALL ((__force gfp_t)___GFP_HARDWALL) /* Enforce hardwall cpuset memory allocs */
87#define __GFP_THISNODE ((__force gfp_t)___GFP_THISNODE)/* No fallback, no policies */
88#define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE) /* Page is reclaimable */
89#define __GFP_NOTRACK ((__force gfp_t)___GFP_NOTRACK) /* Don't track with kmemcheck */
90
91#define __GFP_NO_KSWAPD ((__force gfp_t)___GFP_NO_KSWAPD)
92#define __GFP_OTHER_NODE ((__force gfp_t)___GFP_OTHER_NODE) /* On behalf of other node */
93#define __GFP_KMEMCG ((__force gfp_t)___GFP_KMEMCG) /* Allocation comes from a memcg-accounted resource */
94#define __GFP_WRITE ((__force gfp_t)___GFP_WRITE) /* Allocator intends to dirty page */
95
96/*
97 * This may seem redundant, but it's a way of annotating false positives vs.
98 * allocations that simply cannot be supported (e.g. page tables).
99 */
100#define __GFP_NOTRACK_FALSE_POSITIVE (__GFP_NOTRACK)
101
102#define __GFP_BITS_SHIFT 25 /* Room for N __GFP_FOO bits */
103#define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
104
105/* This equals 0, but use constants in case they ever change */
106#define GFP_NOWAIT (GFP_ATOMIC & ~__GFP_HIGH)
107/* GFP_ATOMIC means both !wait (__GFP_WAIT not set) and use emergency pool */
108#define GFP_ATOMIC (__GFP_HIGH)
109#define GFP_NOIO (__GFP_WAIT)
110#define GFP_NOFS (__GFP_WAIT | __GFP_IO)
111#define GFP_KERNEL (__GFP_WAIT | __GFP_IO | __GFP_FS)
112#define GFP_TEMPORARY (__GFP_WAIT | __GFP_IO | __GFP_FS | \
113 __GFP_RECLAIMABLE)
114#define GFP_USER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
115#define GFP_HIGHUSER (__GFP_WAIT | __GFP_IO | __GFP_FS | __GFP_HARDWALL | \
116 __GFP_HIGHMEM)
117#define GFP_HIGHUSER_MOVABLE (__GFP_WAIT | __GFP_IO | __GFP_FS | \
118 __GFP_HARDWALL | __GFP_HIGHMEM | \
119 __GFP_MOVABLE)
120#define GFP_IOFS (__GFP_IO | __GFP_FS)
121#define GFP_TRANSHUGE (GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
122 __GFP_NOMEMALLOC | __GFP_NORETRY | __GFP_NOWARN | \
123 __GFP_NO_KSWAPD)
124
125#ifdef CONFIG_NUMA
126#define GFP_THISNODE (__GFP_THISNODE | __GFP_NOWARN | __GFP_NORETRY)
127#else
128#define GFP_THISNODE ((__force gfp_t)0)
129#endif
130
131/* This mask makes up all the page movable related flags */
132#define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
133
134/* Control page allocator reclaim behavior */
135#define GFP_RECLAIM_MASK (__GFP_WAIT|__GFP_HIGH|__GFP_IO|__GFP_FS|\
136 __GFP_NOWARN|__GFP_REPEAT|__GFP_NOFAIL|\
137 __GFP_NORETRY|__GFP_MEMALLOC|__GFP_NOMEMALLOC)
138
139/* Control slab gfp mask during early boot */
140#define GFP_BOOT_MASK (__GFP_BITS_MASK & ~(__GFP_WAIT|__GFP_IO|__GFP_FS))
141
142/* Control allocation constraints */
143#define GFP_CONSTRAINT_MASK (__GFP_HARDWALL|__GFP_THISNODE)
144
145/* Do not use these with a slab allocator */
146#define GFP_SLAB_BUG_MASK (__GFP_DMA32|__GFP_HIGHMEM|~__GFP_BITS_MASK)
147
148/* Flag - indicates that the buffer will be suitable for DMA. Ignored on some
149 platforms, used as appropriate on others */
150
151#define GFP_DMA __GFP_DMA
152
153/* 4GB DMA on some platforms */
154#define GFP_DMA32 __GFP_DMA32
155
156/* Convert GFP flags to their corresponding migrate type */
157static inline int allocflags_to_migratetype(gfp_t gfp_flags)
158{
159 WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
160
161 if (unlikely(page_group_by_mobility_disabled))
162 return MIGRATE_UNMOVABLE;
163
164 /* Group based on mobility */
165 return (((gfp_flags & __GFP_MOVABLE) != 0) << 1) |
166 ((gfp_flags & __GFP_RECLAIMABLE) != 0);
167}
168
169#ifdef CONFIG_HIGHMEM
170#define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
171#else
172#define OPT_ZONE_HIGHMEM ZONE_NORMAL
173#endif
174
175#ifdef CONFIG_ZONE_DMA
176#define OPT_ZONE_DMA ZONE_DMA
177#else
178#define OPT_ZONE_DMA ZONE_NORMAL
179#endif
180
181#ifdef CONFIG_ZONE_DMA32
182#define OPT_ZONE_DMA32 ZONE_DMA32
183#else
184#define OPT_ZONE_DMA32 ZONE_NORMAL
185#endif
186
187/*
188 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
189 * zone to use given the lowest 4 bits of gfp_t. Entries are ZONE_SHIFT long
190 * and there are 16 of them to cover all possible combinations of
191 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
192 *
193 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
194 * But GFP_MOVABLE is not only a zone specifier but also an allocation
195 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
196 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
197 *
198 * bit result
199 * =================
200 * 0x0 => NORMAL
201 * 0x1 => DMA or NORMAL
202 * 0x2 => HIGHMEM or NORMAL
203 * 0x3 => BAD (DMA+HIGHMEM)
204 * 0x4 => DMA32 or DMA or NORMAL
205 * 0x5 => BAD (DMA+DMA32)
206 * 0x6 => BAD (HIGHMEM+DMA32)
207 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
208 * 0x8 => NORMAL (MOVABLE+0)
209 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
210 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
211 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
212 * 0xc => DMA32 (MOVABLE+DMA32)
213 * 0xd => BAD (MOVABLE+DMA32+DMA)
214 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
215 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
216 *
217 * ZONES_SHIFT must be <= 2 on 32 bit platforms.
218 */
219
220#if 16 * ZONES_SHIFT > BITS_PER_LONG
221#error ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
222#endif
223
224#define GFP_ZONE_TABLE ( \
225 (ZONE_NORMAL << 0 * ZONES_SHIFT) \
226 | (OPT_ZONE_DMA << ___GFP_DMA * ZONES_SHIFT) \
227 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * ZONES_SHIFT) \
228 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * ZONES_SHIFT) \
229 | (ZONE_NORMAL << ___GFP_MOVABLE * ZONES_SHIFT) \
230 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * ZONES_SHIFT) \
231 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * ZONES_SHIFT) \
232 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * ZONES_SHIFT) \
233)
234
235/*
236 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
237 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
238 * entry starting with bit 0. Bit is set if the combination is not
239 * allowed.
240 */
241#define GFP_ZONE_BAD ( \
242 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
243 | 1 << (___GFP_DMA | ___GFP_DMA32) \
244 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
245 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
246 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
247 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
248 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
249 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
250)
251
252static inline enum zone_type gfp_zone(gfp_t flags)
253{
254 enum zone_type z;
255 int bit = (__force int) (flags & GFP_ZONEMASK);
256
257 z = (GFP_ZONE_TABLE >> (bit * ZONES_SHIFT)) &
258 ((1 << ZONES_SHIFT) - 1);
259 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
260 return z;
261}
262
263/*
264 * There is only one page-allocator function, and two main namespaces to
265 * it. The alloc_page*() variants return 'struct page *' and as such
266 * can allocate highmem pages, the *get*page*() variants return
267 * virtual kernel addresses to the allocated page(s).
268 */
269
270static inline int gfp_zonelist(gfp_t flags)
271{
272 if (IS_ENABLED(CONFIG_NUMA) && unlikely(flags & __GFP_THISNODE))
273 return 1;
274
275 return 0;
276}
277
278/*
279 * We get the zone list from the current node and the gfp_mask.
280 * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
281 * There are two zonelists per node, one for all zones with memory and
282 * one containing just zones from the node the zonelist belongs to.
283 *
284 * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
285 * optimized to &contig_page_data at compile-time.
286 */
287static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
288{
289 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
290}
291
292#ifndef HAVE_ARCH_FREE_PAGE
293static inline void arch_free_page(struct page *page, int order) { }
294#endif
295#ifndef HAVE_ARCH_ALLOC_PAGE
296static inline void arch_alloc_page(struct page *page, int order) { }
297#endif
298
299struct page *
300__alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order,
301 struct zonelist *zonelist, nodemask_t *nodemask);
302
303static inline struct page *
304__alloc_pages(gfp_t gfp_mask, unsigned int order,
305 struct zonelist *zonelist)
306{
307 return __alloc_pages_nodemask(gfp_mask, order, zonelist, NULL);
308}
309
310static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
311 unsigned int order)
312{
313 /* Unknown node is current node */
314 if (nid < 0)
315 nid = numa_node_id();
316
317 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
318}
319
320static inline struct page *alloc_pages_exact_node(int nid, gfp_t gfp_mask,
321 unsigned int order)
322{
323 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES || !node_online(nid));
324
325 return __alloc_pages(gfp_mask, order, node_zonelist(nid, gfp_mask));
326}
327
328#ifdef CONFIG_NUMA
329extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
330
331static inline struct page *
332alloc_pages(gfp_t gfp_mask, unsigned int order)
333{
334 return alloc_pages_current(gfp_mask, order);
335}
336extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
337 struct vm_area_struct *vma, unsigned long addr,
338 int node);
339#else
340#define alloc_pages(gfp_mask, order) \
341 alloc_pages_node(numa_node_id(), gfp_mask, order)
342#define alloc_pages_vma(gfp_mask, order, vma, addr, node) \
343 alloc_pages(gfp_mask, order)
344#endif
345#define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
346#define alloc_page_vma(gfp_mask, vma, addr) \
347 alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id())
348#define alloc_page_vma_node(gfp_mask, vma, addr, node) \
349 alloc_pages_vma(gfp_mask, 0, vma, addr, node)
350
351extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
352extern unsigned long get_zeroed_page(gfp_t gfp_mask);
353
354void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
355void free_pages_exact(void *virt, size_t size);
356/* This is different from alloc_pages_exact_node !!! */
357void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
358
359#define __get_free_page(gfp_mask) \
360 __get_free_pages((gfp_mask), 0)
361
362#define __get_dma_pages(gfp_mask, order) \
363 __get_free_pages((gfp_mask) | GFP_DMA, (order))
364
365extern void __free_pages(struct page *page, unsigned int order);
366extern void free_pages(unsigned long addr, unsigned int order);
367extern void free_hot_cold_page(struct page *page, int cold);
368extern void free_hot_cold_page_list(struct list_head *list, int cold);
369
370extern void __free_memcg_kmem_pages(struct page *page, unsigned int order);
371extern void free_memcg_kmem_pages(unsigned long addr, unsigned int order);
372
373#define __free_page(page) __free_pages((page), 0)
374#define free_page(addr) free_pages((addr), 0)
375
376void page_alloc_init(void);
377void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
378void drain_all_pages(void);
379void drain_local_pages(void *dummy);
380
381/*
382 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
383 * GFP flags are used before interrupts are enabled. Once interrupts are
384 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
385 * hibernation, it is used by PM to avoid I/O during memory allocation while
386 * devices are suspended.
387 */
388extern gfp_t gfp_allowed_mask;
389
390/* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
391bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
392
393extern void pm_restrict_gfp_mask(void);
394extern void pm_restore_gfp_mask(void);
395
396#ifdef CONFIG_PM_SLEEP
397extern bool pm_suspended_storage(void);
398#else
399static inline bool pm_suspended_storage(void)
400{
401 return false;
402}
403#endif /* CONFIG_PM_SLEEP */
404
405#ifdef CONFIG_CMA
406
407/* The below functions must be run on a range from a single zone. */
408extern int alloc_contig_range(unsigned long start, unsigned long end,
409 unsigned migratetype);
410extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
411
412/* CMA stuff */
413extern void init_cma_reserved_pageblock(struct page *page);
414
415#endif
416
417#endif /* __LINUX_GFP_H */