1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef __LINUX_GFP_H
3 #define __LINUX_GFP_H
4 
5 #include <linux/mmdebug.h>
6 #include <linux/mmzone.h>
7 #include <linux/stddef.h>
8 #include <linux/linkage.h>
9 #include <linux/topology.h>
10 
11 struct vm_area_struct;
12 
13 /*
14  * In case of changes, please don't forget to update
15  * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
16  */
17 
18 /* Plain integer GFP bitmasks. Do not use this directly. */
19 #define ___GFP_DMA		0x01u
20 #define ___GFP_HIGHMEM		0x02u
21 #define ___GFP_DMA32		0x04u
22 #define ___GFP_MOVABLE		0x08u
23 #define ___GFP_RECLAIMABLE	0x10u
24 #define ___GFP_HIGH		0x20u
25 #define ___GFP_IO		0x40u
26 #define ___GFP_FS		0x80u
27 #define ___GFP_WRITE		0x100u
28 #define ___GFP_NOWARN		0x200u
29 #define ___GFP_RETRY_MAYFAIL	0x400u
30 #define ___GFP_NOFAIL		0x800u
31 #define ___GFP_NORETRY		0x1000u
32 #define ___GFP_MEMALLOC		0x2000u
33 #define ___GFP_COMP		0x4000u
34 #define ___GFP_ZERO		0x8000u
35 #define ___GFP_NOMEMALLOC	0x10000u
36 #define ___GFP_HARDWALL		0x20000u
37 #define ___GFP_THISNODE		0x40000u
38 #define ___GFP_ATOMIC		0x80000u
39 #define ___GFP_ACCOUNT		0x100000u
40 #define ___GFP_DIRECT_RECLAIM	0x200000u
41 #define ___GFP_KSWAPD_RECLAIM	0x400000u
42 #ifdef CONFIG_LOCKDEP
43 #define ___GFP_NOLOCKDEP	0x800000u
44 #else
45 #define ___GFP_NOLOCKDEP	0
46 #endif
47 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
48 
49 /*
50  * Physical address zone modifiers (see linux/mmzone.h - low four bits)
51  *
52  * Do not put any conditional on these. If necessary modify the definitions
53  * without the underscores and use them consistently. The definitions here may
54  * be used in bit comparisons.
55  */
56 #define __GFP_DMA	((__force gfp_t)___GFP_DMA)
57 #define __GFP_HIGHMEM	((__force gfp_t)___GFP_HIGHMEM)
58 #define __GFP_DMA32	((__force gfp_t)___GFP_DMA32)
59 #define __GFP_MOVABLE	((__force gfp_t)___GFP_MOVABLE)  /* ZONE_MOVABLE allowed */
60 #define GFP_ZONEMASK	(__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
61 
62 /**
63  * DOC: Page mobility and placement hints
64  *
65  * Page mobility and placement hints
66  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
67  *
68  * These flags provide hints about how mobile the page is. Pages with similar
69  * mobility are placed within the same pageblocks to minimise problems due
70  * to external fragmentation.
71  *
72  * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be
73  * moved by page migration during memory compaction or can be reclaimed.
74  *
75  * %__GFP_RECLAIMABLE is used for slab allocations that specify
76  * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
77  *
78  * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible,
79  * these pages will be spread between local zones to avoid all the dirty
80  * pages being in one zone (fair zone allocation policy).
81  *
82  * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
83  *
84  * %__GFP_THISNODE forces the allocation to be satisified from the requested
85  * node with no fallbacks or placement policy enforcements.
86  *
87  * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
88  */
89 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
90 #define __GFP_WRITE	((__force gfp_t)___GFP_WRITE)
91 #define __GFP_HARDWALL   ((__force gfp_t)___GFP_HARDWALL)
92 #define __GFP_THISNODE	((__force gfp_t)___GFP_THISNODE)
93 #define __GFP_ACCOUNT	((__force gfp_t)___GFP_ACCOUNT)
94 
95 /**
96  * DOC: Watermark modifiers
97  *
98  * Watermark modifiers -- controls access to emergency reserves
99  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
100  *
101  * %__GFP_HIGH indicates that the caller is high-priority and that granting
102  * the request is necessary before the system can make forward progress.
103  * For example, creating an IO context to clean pages.
104  *
105  * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
106  * high priority. Users are typically interrupt handlers. This may be
107  * used in conjunction with %__GFP_HIGH
108  *
109  * %__GFP_MEMALLOC allows access to all memory. This should only be used when
110  * the caller guarantees the allocation will allow more memory to be freed
111  * very shortly e.g. process exiting or swapping. Users either should
112  * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
113  *
114  * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
115  * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
116  */
117 #define __GFP_ATOMIC	((__force gfp_t)___GFP_ATOMIC)
118 #define __GFP_HIGH	((__force gfp_t)___GFP_HIGH)
119 #define __GFP_MEMALLOC	((__force gfp_t)___GFP_MEMALLOC)
120 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
121 
122 /**
123  * DOC: Reclaim modifiers
124  *
125  * Reclaim modifiers
126  * ~~~~~~~~~~~~~~~~~
127  *
128  * %__GFP_IO can start physical IO.
129  *
130  * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
131  * allocator recursing into the filesystem which might already be holding
132  * locks.
133  *
134  * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
135  * This flag can be cleared to avoid unnecessary delays when a fallback
136  * option is available.
137  *
138  * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
139  * the low watermark is reached and have it reclaim pages until the high
140  * watermark is reached. A caller may wish to clear this flag when fallback
141  * options are available and the reclaim is likely to disrupt the system. The
142  * canonical example is THP allocation where a fallback is cheap but
143  * reclaim/compaction may cause indirect stalls.
144  *
145  * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
146  *
147  * The default allocator behavior depends on the request size. We have a concept
148  * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
149  * !costly allocations are too essential to fail so they are implicitly
150  * non-failing by default (with some exceptions like OOM victims might fail so
151  * the caller still has to check for failures) while costly requests try to be
152  * not disruptive and back off even without invoking the OOM killer.
153  * The following three modifiers might be used to override some of these
154  * implicit rules
155  *
156  * %__GFP_NORETRY: The VM implementation will try only very lightweight
157  * memory direct reclaim to get some memory under memory pressure (thus
158  * it can sleep). It will avoid disruptive actions like OOM killer. The
159  * caller must handle the failure which is quite likely to happen under
160  * heavy memory pressure. The flag is suitable when failure can easily be
161  * handled at small cost, such as reduced throughput
162  *
163  * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
164  * procedures that have previously failed if there is some indication
165  * that progress has been made else where.  It can wait for other
166  * tasks to attempt high level approaches to freeing memory such as
167  * compaction (which removes fragmentation) and page-out.
168  * There is still a definite limit to the number of retries, but it is
169  * a larger limit than with %__GFP_NORETRY.
170  * Allocations with this flag may fail, but only when there is
171  * genuinely little unused memory. While these allocations do not
172  * directly trigger the OOM killer, their failure indicates that
173  * the system is likely to need to use the OOM killer soon.  The
174  * caller must handle failure, but can reasonably do so by failing
175  * a higher-level request, or completing it only in a much less
176  * efficient manner.
177  * If the allocation does fail, and the caller is in a position to
178  * free some non-essential memory, doing so could benefit the system
179  * as a whole.
180  *
181  * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
182  * cannot handle allocation failures. The allocation could block
183  * indefinitely but will never return with failure. Testing for
184  * failure is pointless.
185  * New users should be evaluated carefully (and the flag should be
186  * used only when there is no reasonable failure policy) but it is
187  * definitely preferable to use the flag rather than opencode endless
188  * loop around allocator.
189  * Using this flag for costly allocations is _highly_ discouraged.
190  */
191 #define __GFP_IO	((__force gfp_t)___GFP_IO)
192 #define __GFP_FS	((__force gfp_t)___GFP_FS)
193 #define __GFP_DIRECT_RECLAIM	((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
194 #define __GFP_KSWAPD_RECLAIM	((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
195 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
196 #define __GFP_RETRY_MAYFAIL	((__force gfp_t)___GFP_RETRY_MAYFAIL)
197 #define __GFP_NOFAIL	((__force gfp_t)___GFP_NOFAIL)
198 #define __GFP_NORETRY	((__force gfp_t)___GFP_NORETRY)
199 
200 /**
201  * DOC: Action modifiers
202  *
203  * Action modifiers
204  * ~~~~~~~~~~~~~~~~
205  *
206  * %__GFP_NOWARN suppresses allocation failure reports.
207  *
208  * %__GFP_COMP address compound page metadata.
209  *
210  * %__GFP_ZERO returns a zeroed page on success.
211  */
212 #define __GFP_NOWARN	((__force gfp_t)___GFP_NOWARN)
213 #define __GFP_COMP	((__force gfp_t)___GFP_COMP)
214 #define __GFP_ZERO	((__force gfp_t)___GFP_ZERO)
215 
216 /* Disable lockdep for GFP context tracking */
217 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
218 
219 /* Room for N __GFP_FOO bits */
220 #define __GFP_BITS_SHIFT (23 + IS_ENABLED(CONFIG_LOCKDEP))
221 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
222 
223 /**
224  * DOC: Useful GFP flag combinations
225  *
226  * Useful GFP flag combinations
227  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
228  *
229  * Useful GFP flag combinations that are commonly used. It is recommended
230  * that subsystems start with one of these combinations and then set/clear
231  * %__GFP_FOO flags as necessary.
232  *
233  * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
234  * watermark is applied to allow access to "atomic reserves"
235  *
236  * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
237  * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
238  *
239  * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
240  * accounted to kmemcg.
241  *
242  * %GFP_NOWAIT is for kernel allocations that should not stall for direct
243  * reclaim, start physical IO or use any filesystem callback.
244  *
245  * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
246  * that do not require the starting of any physical IO.
247  * Please try to avoid using this flag directly and instead use
248  * memalloc_noio_{save,restore} to mark the whole scope which cannot
249  * perform any IO with a short explanation why. All allocation requests
250  * will inherit GFP_NOIO implicitly.
251  *
252  * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
253  * Please try to avoid using this flag directly and instead use
254  * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
255  * recurse into the FS layer with a short explanation why. All allocation
256  * requests will inherit GFP_NOFS implicitly.
257  *
258  * %GFP_USER is for userspace allocations that also need to be directly
259  * accessibly by the kernel or hardware. It is typically used by hardware
260  * for buffers that are mapped to userspace (e.g. graphics) that hardware
261  * still must DMA to. cpuset limits are enforced for these allocations.
262  *
263  * %GFP_DMA exists for historical reasons and should be avoided where possible.
264  * The flags indicates that the caller requires that the lowest zone be
265  * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
266  * it would require careful auditing as some users really require it and
267  * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
268  * lowest zone as a type of emergency reserve.
269  *
270  * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
271  * address.
272  *
273  * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
274  * do not need to be directly accessible by the kernel but that cannot
275  * move once in use. An example may be a hardware allocation that maps
276  * data directly into userspace but has no addressing limitations.
277  *
278  * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
279  * need direct access to but can use kmap() when access is required. They
280  * are expected to be movable via page reclaim or page migration. Typically,
281  * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
282  *
283  * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
284  * are compound allocations that will generally fail quickly if memory is not
285  * available and will not wake kswapd/kcompactd on failure. The _LIGHT
286  * version does not attempt reclaim/compaction at all and is by default used
287  * in page fault path, while the non-light is used by khugepaged.
288  */
289 #define GFP_ATOMIC	(__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
290 #define GFP_KERNEL	(__GFP_RECLAIM | __GFP_IO | __GFP_FS)
291 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
292 #define GFP_NOWAIT	(__GFP_KSWAPD_RECLAIM)
293 #define GFP_NOIO	(__GFP_RECLAIM)
294 #define GFP_NOFS	(__GFP_RECLAIM | __GFP_IO)
295 #define GFP_USER	(__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
296 #define GFP_DMA		__GFP_DMA
297 #define GFP_DMA32	__GFP_DMA32
298 #define GFP_HIGHUSER	(GFP_USER | __GFP_HIGHMEM)
299 #define GFP_HIGHUSER_MOVABLE	(GFP_HIGHUSER | __GFP_MOVABLE)
300 #define GFP_TRANSHUGE_LIGHT	((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
301 			 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
302 #define GFP_TRANSHUGE	(GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
303 
304 /* Convert GFP flags to their corresponding migrate type */
305 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
306 #define GFP_MOVABLE_SHIFT 3
307 
gfpflags_to_migratetype(const gfp_t gfp_flags)308 static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
309 {
310 	VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
311 	BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
312 	BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
313 
314 	if (unlikely(page_group_by_mobility_disabled))
315 		return MIGRATE_UNMOVABLE;
316 
317 	/* Group based on mobility */
318 	return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
319 }
320 #undef GFP_MOVABLE_MASK
321 #undef GFP_MOVABLE_SHIFT
322 
gfpflags_allow_blocking(const gfp_t gfp_flags)323 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
324 {
325 	return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
326 }
327 
328 /**
329  * gfpflags_normal_context - is gfp_flags a normal sleepable context?
330  * @gfp_flags: gfp_flags to test
331  *
332  * Test whether @gfp_flags indicates that the allocation is from the
333  * %current context and allowed to sleep.
334  *
335  * An allocation being allowed to block doesn't mean it owns the %current
336  * context.  When direct reclaim path tries to allocate memory, the
337  * allocation context is nested inside whatever %current was doing at the
338  * time of the original allocation.  The nested allocation may be allowed
339  * to block but modifying anything %current owns can corrupt the outer
340  * context's expectations.
341  *
342  * %true result from this function indicates that the allocation context
343  * can sleep and use anything that's associated with %current.
344  */
gfpflags_normal_context(const gfp_t gfp_flags)345 static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
346 {
347 	return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
348 		__GFP_DIRECT_RECLAIM;
349 }
350 
351 #ifdef CONFIG_HIGHMEM
352 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
353 #else
354 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
355 #endif
356 
357 #ifdef CONFIG_ZONE_DMA
358 #define OPT_ZONE_DMA ZONE_DMA
359 #else
360 #define OPT_ZONE_DMA ZONE_NORMAL
361 #endif
362 
363 #ifdef CONFIG_ZONE_DMA32
364 #define OPT_ZONE_DMA32 ZONE_DMA32
365 #else
366 #define OPT_ZONE_DMA32 ZONE_NORMAL
367 #endif
368 
369 /*
370  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
371  * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
372  * bits long and there are 16 of them to cover all possible combinations of
373  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
374  *
375  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
376  * But GFP_MOVABLE is not only a zone specifier but also an allocation
377  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
378  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
379  *
380  *       bit       result
381  *       =================
382  *       0x0    => NORMAL
383  *       0x1    => DMA or NORMAL
384  *       0x2    => HIGHMEM or NORMAL
385  *       0x3    => BAD (DMA+HIGHMEM)
386  *       0x4    => DMA32 or NORMAL
387  *       0x5    => BAD (DMA+DMA32)
388  *       0x6    => BAD (HIGHMEM+DMA32)
389  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
390  *       0x8    => NORMAL (MOVABLE+0)
391  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
392  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
393  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
394  *       0xc    => DMA32 or NORMAL (MOVABLE+DMA32)
395  *       0xd    => BAD (MOVABLE+DMA32+DMA)
396  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
397  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
398  *
399  * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
400  */
401 
402 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
403 /* ZONE_DEVICE is not a valid GFP zone specifier */
404 #define GFP_ZONES_SHIFT 2
405 #else
406 #define GFP_ZONES_SHIFT ZONES_SHIFT
407 #endif
408 
409 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
410 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
411 #endif
412 
413 #define GFP_ZONE_TABLE ( \
414 	(ZONE_NORMAL << 0 * GFP_ZONES_SHIFT)				       \
415 	| (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT)		       \
416 	| (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT)	       \
417 	| (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT)		       \
418 	| (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT)		       \
419 	| (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT)    \
420 	| (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
421 	| (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
422 )
423 
424 /*
425  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
426  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
427  * entry starting with bit 0. Bit is set if the combination is not
428  * allowed.
429  */
430 #define GFP_ZONE_BAD ( \
431 	1 << (___GFP_DMA | ___GFP_HIGHMEM)				      \
432 	| 1 << (___GFP_DMA | ___GFP_DMA32)				      \
433 	| 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)				      \
434 	| 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
435 	| 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)		      \
436 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)		      \
437 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
438 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
439 )
440 
gfp_zone(gfp_t flags)441 static inline enum zone_type gfp_zone(gfp_t flags)
442 {
443 	enum zone_type z;
444 	int bit = (__force int) (flags & GFP_ZONEMASK);
445 
446 	z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
447 					 ((1 << GFP_ZONES_SHIFT) - 1);
448 	VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
449 	return z;
450 }
451 
452 /*
453  * There is only one page-allocator function, and two main namespaces to
454  * it. The alloc_page*() variants return 'struct page *' and as such
455  * can allocate highmem pages, the *get*page*() variants return
456  * virtual kernel addresses to the allocated page(s).
457  */
458 
gfp_zonelist(gfp_t flags)459 static inline int gfp_zonelist(gfp_t flags)
460 {
461 #ifdef CONFIG_NUMA
462 	if (unlikely(flags & __GFP_THISNODE))
463 		return ZONELIST_NOFALLBACK;
464 #endif
465 	return ZONELIST_FALLBACK;
466 }
467 
468 /*
469  * We get the zone list from the current node and the gfp_mask.
470  * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
471  * There are two zonelists per node, one for all zones with memory and
472  * one containing just zones from the node the zonelist belongs to.
473  *
474  * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
475  * optimized to &contig_page_data at compile-time.
476  */
node_zonelist(int nid,gfp_t flags)477 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
478 {
479 	return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
480 }
481 
482 #ifndef HAVE_ARCH_FREE_PAGE
arch_free_page(struct page * page,int order)483 static inline void arch_free_page(struct page *page, int order) { }
484 #endif
485 #ifndef HAVE_ARCH_ALLOC_PAGE
arch_alloc_page(struct page * page,int order)486 static inline void arch_alloc_page(struct page *page, int order) { }
487 #endif
488 
489 struct page *
490 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
491 							nodemask_t *nodemask);
492 
493 static inline struct page *
__alloc_pages(gfp_t gfp_mask,unsigned int order,int preferred_nid)494 __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
495 {
496 	return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL);
497 }
498 
499 /*
500  * Allocate pages, preferring the node given as nid. The node must be valid and
501  * online. For more general interface, see alloc_pages_node().
502  */
503 static inline struct page *
__alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)504 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
505 {
506 	VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
507 	VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
508 
509 	return __alloc_pages(gfp_mask, order, nid);
510 }
511 
512 /*
513  * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
514  * prefer the current CPU's closest node. Otherwise node must be valid and
515  * online.
516  */
alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)517 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
518 						unsigned int order)
519 {
520 	if (nid == NUMA_NO_NODE)
521 		nid = numa_mem_id();
522 
523 	return __alloc_pages_node(nid, gfp_mask, order);
524 }
525 
526 #ifdef CONFIG_NUMA
527 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
528 
529 static inline struct page *
alloc_pages(gfp_t gfp_mask,unsigned int order)530 alloc_pages(gfp_t gfp_mask, unsigned int order)
531 {
532 	return alloc_pages_current(gfp_mask, order);
533 }
534 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
535 			struct vm_area_struct *vma, unsigned long addr,
536 			int node, bool hugepage);
537 #define alloc_hugepage_vma(gfp_mask, vma, addr, order)	\
538 	alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
539 #else
540 #define alloc_pages(gfp_mask, order) \
541 		alloc_pages_node(numa_node_id(), gfp_mask, order)
542 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
543 	alloc_pages(gfp_mask, order)
544 #define alloc_hugepage_vma(gfp_mask, vma, addr, order)	\
545 	alloc_pages(gfp_mask, order)
546 #endif
547 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
548 #define alloc_page_vma(gfp_mask, vma, addr)			\
549 	alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
550 #define alloc_page_vma_node(gfp_mask, vma, addr, node)		\
551 	alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
552 
553 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
554 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
555 
556 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
557 void free_pages_exact(void *virt, size_t size);
558 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
559 
560 #define __get_free_page(gfp_mask) \
561 		__get_free_pages((gfp_mask), 0)
562 
563 #define __get_dma_pages(gfp_mask, order) \
564 		__get_free_pages((gfp_mask) | GFP_DMA, (order))
565 
566 extern void __free_pages(struct page *page, unsigned int order);
567 extern void free_pages(unsigned long addr, unsigned int order);
568 extern void free_unref_page(struct page *page);
569 extern void free_unref_page_list(struct list_head *list);
570 
571 struct page_frag_cache;
572 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
573 extern void *page_frag_alloc(struct page_frag_cache *nc,
574 			     unsigned int fragsz, gfp_t gfp_mask);
575 extern void page_frag_free(void *addr);
576 
577 #define __free_page(page) __free_pages((page), 0)
578 #define free_page(addr) free_pages((addr), 0)
579 
580 void page_alloc_init(void);
581 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
582 void drain_all_pages(struct zone *zone);
583 void drain_local_pages(struct zone *zone);
584 
585 void page_alloc_init_late(void);
586 
587 /*
588  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
589  * GFP flags are used before interrupts are enabled. Once interrupts are
590  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
591  * hibernation, it is used by PM to avoid I/O during memory allocation while
592  * devices are suspended.
593  */
594 extern gfp_t gfp_allowed_mask;
595 
596 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
597 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
598 
599 extern void pm_restrict_gfp_mask(void);
600 extern void pm_restore_gfp_mask(void);
601 
602 #ifdef CONFIG_PM_SLEEP
603 extern bool pm_suspended_storage(void);
604 #else
pm_suspended_storage(void)605 static inline bool pm_suspended_storage(void)
606 {
607 	return false;
608 }
609 #endif /* CONFIG_PM_SLEEP */
610 
611 #if (defined(CONFIG_MEMORY_ISOLATION) && defined(CONFIG_COMPACTION)) || defined(CONFIG_CMA)
612 /* The below functions must be run on a range from a single zone. */
613 extern int alloc_contig_range(unsigned long start, unsigned long end,
614 			      unsigned migratetype, gfp_t gfp_mask);
615 extern void free_contig_range(unsigned long pfn, unsigned nr_pages);
616 #endif
617 
618 #ifdef CONFIG_CMA
619 /* CMA stuff */
620 extern void init_cma_reserved_pageblock(struct page *page);
621 #endif
622 
623 #endif /* __LINUX_GFP_H */
624