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 /* The typedef is in types.h but we want the documentation here */
12 #if 0
13 /**
14  * typedef gfp_t - Memory allocation flags.
15  *
16  * GFP flags are commonly used throughout Linux to indicate how memory
17  * should be allocated.  The GFP acronym stands for get_free_pages(),
18  * the underlying memory allocation function.  Not every GFP flag is
19  * supported by every function which may allocate memory.  Most users
20  * will want to use a plain ``GFP_KERNEL``.
21  */
22 typedef unsigned int __bitwise gfp_t;
23 #endif
24 
25 struct vm_area_struct;
26 
27 /*
28  * In case of changes, please don't forget to update
29  * include/trace/events/mmflags.h and tools/perf/builtin-kmem.c
30  */
31 
32 /* Plain integer GFP bitmasks. Do not use this directly. */
33 #define ___GFP_DMA		0x01u
34 #define ___GFP_HIGHMEM		0x02u
35 #define ___GFP_DMA32		0x04u
36 #define ___GFP_MOVABLE		0x08u
37 #define ___GFP_RECLAIMABLE	0x10u
38 #define ___GFP_HIGH		0x20u
39 #define ___GFP_IO		0x40u
40 #define ___GFP_FS		0x80u
41 #define ___GFP_ZERO		0x100u
42 #define ___GFP_ATOMIC		0x200u
43 #define ___GFP_DIRECT_RECLAIM	0x400u
44 #define ___GFP_KSWAPD_RECLAIM	0x800u
45 #define ___GFP_WRITE		0x1000u
46 #define ___GFP_NOWARN		0x2000u
47 #define ___GFP_RETRY_MAYFAIL	0x4000u
48 #define ___GFP_NOFAIL		0x8000u
49 #define ___GFP_NORETRY		0x10000u
50 #define ___GFP_MEMALLOC		0x20000u
51 #define ___GFP_COMP		0x40000u
52 #define ___GFP_NOMEMALLOC	0x80000u
53 #define ___GFP_HARDWALL		0x100000u
54 #define ___GFP_THISNODE		0x200000u
55 #define ___GFP_ACCOUNT		0x400000u
56 #define ___GFP_ZEROTAGS		0x800000u
57 #ifdef CONFIG_KASAN_HW_TAGS
58 #define ___GFP_SKIP_ZERO		0x1000000u
59 #define ___GFP_SKIP_KASAN_UNPOISON	0x2000000u
60 #define ___GFP_SKIP_KASAN_POISON	0x4000000u
61 #else
62 #define ___GFP_SKIP_ZERO		0
63 #define ___GFP_SKIP_KASAN_UNPOISON	0
64 #define ___GFP_SKIP_KASAN_POISON	0
65 #endif
66 #ifdef CONFIG_LOCKDEP
67 #define ___GFP_NOLOCKDEP	0x8000000u
68 #else
69 #define ___GFP_NOLOCKDEP	0
70 #endif
71 /* If the above are modified, __GFP_BITS_SHIFT may need updating */
72 
73 /*
74  * Physical address zone modifiers (see linux/mmzone.h - low four bits)
75  *
76  * Do not put any conditional on these. If necessary modify the definitions
77  * without the underscores and use them consistently. The definitions here may
78  * be used in bit comparisons.
79  */
80 #define __GFP_DMA	((__force gfp_t)___GFP_DMA)
81 #define __GFP_HIGHMEM	((__force gfp_t)___GFP_HIGHMEM)
82 #define __GFP_DMA32	((__force gfp_t)___GFP_DMA32)
83 #define __GFP_MOVABLE	((__force gfp_t)___GFP_MOVABLE)  /* ZONE_MOVABLE allowed */
84 #define GFP_ZONEMASK	(__GFP_DMA|__GFP_HIGHMEM|__GFP_DMA32|__GFP_MOVABLE)
85 
86 /**
87  * DOC: Page mobility and placement hints
88  *
89  * Page mobility and placement hints
90  * ---------------------------------
91  *
92  * These flags provide hints about how mobile the page is. Pages with similar
93  * mobility are placed within the same pageblocks to minimise problems due
94  * to external fragmentation.
95  *
96  * %__GFP_MOVABLE (also a zone modifier) indicates that the page can be
97  * moved by page migration during memory compaction or can be reclaimed.
98  *
99  * %__GFP_RECLAIMABLE is used for slab allocations that specify
100  * SLAB_RECLAIM_ACCOUNT and whose pages can be freed via shrinkers.
101  *
102  * %__GFP_WRITE indicates the caller intends to dirty the page. Where possible,
103  * these pages will be spread between local zones to avoid all the dirty
104  * pages being in one zone (fair zone allocation policy).
105  *
106  * %__GFP_HARDWALL enforces the cpuset memory allocation policy.
107  *
108  * %__GFP_THISNODE forces the allocation to be satisfied from the requested
109  * node with no fallbacks or placement policy enforcements.
110  *
111  * %__GFP_ACCOUNT causes the allocation to be accounted to kmemcg.
112  */
113 #define __GFP_RECLAIMABLE ((__force gfp_t)___GFP_RECLAIMABLE)
114 #define __GFP_WRITE	((__force gfp_t)___GFP_WRITE)
115 #define __GFP_HARDWALL   ((__force gfp_t)___GFP_HARDWALL)
116 #define __GFP_THISNODE	((__force gfp_t)___GFP_THISNODE)
117 #define __GFP_ACCOUNT	((__force gfp_t)___GFP_ACCOUNT)
118 
119 /**
120  * DOC: Watermark modifiers
121  *
122  * Watermark modifiers -- controls access to emergency reserves
123  * ------------------------------------------------------------
124  *
125  * %__GFP_HIGH indicates that the caller is high-priority and that granting
126  * the request is necessary before the system can make forward progress.
127  * For example, creating an IO context to clean pages.
128  *
129  * %__GFP_ATOMIC indicates that the caller cannot reclaim or sleep and is
130  * high priority. Users are typically interrupt handlers. This may be
131  * used in conjunction with %__GFP_HIGH
132  *
133  * %__GFP_MEMALLOC allows access to all memory. This should only be used when
134  * the caller guarantees the allocation will allow more memory to be freed
135  * very shortly e.g. process exiting or swapping. Users either should
136  * be the MM or co-ordinating closely with the VM (e.g. swap over NFS).
137  * Users of this flag have to be extremely careful to not deplete the reserve
138  * completely and implement a throttling mechanism which controls the
139  * consumption of the reserve based on the amount of freed memory.
140  * Usage of a pre-allocated pool (e.g. mempool) should be always considered
141  * before using this flag.
142  *
143  * %__GFP_NOMEMALLOC is used to explicitly forbid access to emergency reserves.
144  * This takes precedence over the %__GFP_MEMALLOC flag if both are set.
145  */
146 #define __GFP_ATOMIC	((__force gfp_t)___GFP_ATOMIC)
147 #define __GFP_HIGH	((__force gfp_t)___GFP_HIGH)
148 #define __GFP_MEMALLOC	((__force gfp_t)___GFP_MEMALLOC)
149 #define __GFP_NOMEMALLOC ((__force gfp_t)___GFP_NOMEMALLOC)
150 
151 /**
152  * DOC: Reclaim modifiers
153  *
154  * Reclaim modifiers
155  * -----------------
156  * Please note that all the following flags are only applicable to sleepable
157  * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
158  *
159  * %__GFP_IO can start physical IO.
160  *
161  * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
162  * allocator recursing into the filesystem which might already be holding
163  * locks.
164  *
165  * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
166  * This flag can be cleared to avoid unnecessary delays when a fallback
167  * option is available.
168  *
169  * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
170  * the low watermark is reached and have it reclaim pages until the high
171  * watermark is reached. A caller may wish to clear this flag when fallback
172  * options are available and the reclaim is likely to disrupt the system. The
173  * canonical example is THP allocation where a fallback is cheap but
174  * reclaim/compaction may cause indirect stalls.
175  *
176  * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
177  *
178  * The default allocator behavior depends on the request size. We have a concept
179  * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
180  * !costly allocations are too essential to fail so they are implicitly
181  * non-failing by default (with some exceptions like OOM victims might fail so
182  * the caller still has to check for failures) while costly requests try to be
183  * not disruptive and back off even without invoking the OOM killer.
184  * The following three modifiers might be used to override some of these
185  * implicit rules
186  *
187  * %__GFP_NORETRY: The VM implementation will try only very lightweight
188  * memory direct reclaim to get some memory under memory pressure (thus
189  * it can sleep). It will avoid disruptive actions like OOM killer. The
190  * caller must handle the failure which is quite likely to happen under
191  * heavy memory pressure. The flag is suitable when failure can easily be
192  * handled at small cost, such as reduced throughput
193  *
194  * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
195  * procedures that have previously failed if there is some indication
196  * that progress has been made else where.  It can wait for other
197  * tasks to attempt high level approaches to freeing memory such as
198  * compaction (which removes fragmentation) and page-out.
199  * There is still a definite limit to the number of retries, but it is
200  * a larger limit than with %__GFP_NORETRY.
201  * Allocations with this flag may fail, but only when there is
202  * genuinely little unused memory. While these allocations do not
203  * directly trigger the OOM killer, their failure indicates that
204  * the system is likely to need to use the OOM killer soon.  The
205  * caller must handle failure, but can reasonably do so by failing
206  * a higher-level request, or completing it only in a much less
207  * efficient manner.
208  * If the allocation does fail, and the caller is in a position to
209  * free some non-essential memory, doing so could benefit the system
210  * as a whole.
211  *
212  * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
213  * cannot handle allocation failures. The allocation could block
214  * indefinitely but will never return with failure. Testing for
215  * failure is pointless.
216  * New users should be evaluated carefully (and the flag should be
217  * used only when there is no reasonable failure policy) but it is
218  * definitely preferable to use the flag rather than opencode endless
219  * loop around allocator.
220  * Using this flag for costly allocations is _highly_ discouraged.
221  */
222 #define __GFP_IO	((__force gfp_t)___GFP_IO)
223 #define __GFP_FS	((__force gfp_t)___GFP_FS)
224 #define __GFP_DIRECT_RECLAIM	((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
225 #define __GFP_KSWAPD_RECLAIM	((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
226 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
227 #define __GFP_RETRY_MAYFAIL	((__force gfp_t)___GFP_RETRY_MAYFAIL)
228 #define __GFP_NOFAIL	((__force gfp_t)___GFP_NOFAIL)
229 #define __GFP_NORETRY	((__force gfp_t)___GFP_NORETRY)
230 
231 /**
232  * DOC: Action modifiers
233  *
234  * Action modifiers
235  * ----------------
236  *
237  * %__GFP_NOWARN suppresses allocation failure reports.
238  *
239  * %__GFP_COMP address compound page metadata.
240  *
241  * %__GFP_ZERO returns a zeroed page on success.
242  *
243  * %__GFP_ZEROTAGS zeroes memory tags at allocation time if the memory itself
244  * is being zeroed (either via __GFP_ZERO or via init_on_alloc, provided that
245  * __GFP_SKIP_ZERO is not set). This flag is intended for optimization: setting
246  * memory tags at the same time as zeroing memory has minimal additional
247  * performace impact.
248  *
249  * %__GFP_SKIP_KASAN_UNPOISON makes KASAN skip unpoisoning on page allocation.
250  * Only effective in HW_TAGS mode.
251  *
252  * %__GFP_SKIP_KASAN_POISON makes KASAN skip poisoning on page deallocation.
253  * Typically, used for userspace pages. Only effective in HW_TAGS mode.
254  */
255 #define __GFP_NOWARN	((__force gfp_t)___GFP_NOWARN)
256 #define __GFP_COMP	((__force gfp_t)___GFP_COMP)
257 #define __GFP_ZERO	((__force gfp_t)___GFP_ZERO)
258 #define __GFP_ZEROTAGS	((__force gfp_t)___GFP_ZEROTAGS)
259 #define __GFP_SKIP_ZERO ((__force gfp_t)___GFP_SKIP_ZERO)
260 #define __GFP_SKIP_KASAN_UNPOISON ((__force gfp_t)___GFP_SKIP_KASAN_UNPOISON)
261 #define __GFP_SKIP_KASAN_POISON   ((__force gfp_t)___GFP_SKIP_KASAN_POISON)
262 
263 /* Disable lockdep for GFP context tracking */
264 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
265 
266 /* Room for N __GFP_FOO bits */
267 #define __GFP_BITS_SHIFT (27 + IS_ENABLED(CONFIG_LOCKDEP))
268 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
269 
270 /**
271  * DOC: Useful GFP flag combinations
272  *
273  * Useful GFP flag combinations
274  * ----------------------------
275  *
276  * Useful GFP flag combinations that are commonly used. It is recommended
277  * that subsystems start with one of these combinations and then set/clear
278  * %__GFP_FOO flags as necessary.
279  *
280  * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
281  * watermark is applied to allow access to "atomic reserves".
282  * The current implementation doesn't support NMI and few other strict
283  * non-preemptive contexts (e.g. raw_spin_lock). The same applies to %GFP_NOWAIT.
284  *
285  * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
286  * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
287  *
288  * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
289  * accounted to kmemcg.
290  *
291  * %GFP_NOWAIT is for kernel allocations that should not stall for direct
292  * reclaim, start physical IO or use any filesystem callback.
293  *
294  * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
295  * that do not require the starting of any physical IO.
296  * Please try to avoid using this flag directly and instead use
297  * memalloc_noio_{save,restore} to mark the whole scope which cannot
298  * perform any IO with a short explanation why. All allocation requests
299  * will inherit GFP_NOIO implicitly.
300  *
301  * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
302  * Please try to avoid using this flag directly and instead use
303  * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
304  * recurse into the FS layer with a short explanation why. All allocation
305  * requests will inherit GFP_NOFS implicitly.
306  *
307  * %GFP_USER is for userspace allocations that also need to be directly
308  * accessibly by the kernel or hardware. It is typically used by hardware
309  * for buffers that are mapped to userspace (e.g. graphics) that hardware
310  * still must DMA to. cpuset limits are enforced for these allocations.
311  *
312  * %GFP_DMA exists for historical reasons and should be avoided where possible.
313  * The flags indicates that the caller requires that the lowest zone be
314  * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
315  * it would require careful auditing as some users really require it and
316  * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
317  * lowest zone as a type of emergency reserve.
318  *
319  * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
320  * address. Note that kmalloc(..., GFP_DMA32) does not return DMA32 memory
321  * because the DMA32 kmalloc cache array is not implemented.
322  * (Reason: there is no such user in kernel).
323  *
324  * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
325  * do not need to be directly accessible by the kernel but that cannot
326  * move once in use. An example may be a hardware allocation that maps
327  * data directly into userspace but has no addressing limitations.
328  *
329  * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
330  * need direct access to but can use kmap() when access is required. They
331  * are expected to be movable via page reclaim or page migration. Typically,
332  * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
333  *
334  * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
335  * are compound allocations that will generally fail quickly if memory is not
336  * available and will not wake kswapd/kcompactd on failure. The _LIGHT
337  * version does not attempt reclaim/compaction at all and is by default used
338  * in page fault path, while the non-light is used by khugepaged.
339  */
340 #define GFP_ATOMIC	(__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
341 #define GFP_KERNEL	(__GFP_RECLAIM | __GFP_IO | __GFP_FS)
342 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
343 #define GFP_NOWAIT	(__GFP_KSWAPD_RECLAIM)
344 #define GFP_NOIO	(__GFP_RECLAIM)
345 #define GFP_NOFS	(__GFP_RECLAIM | __GFP_IO)
346 #define GFP_USER	(__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
347 #define GFP_DMA		__GFP_DMA
348 #define GFP_DMA32	__GFP_DMA32
349 #define GFP_HIGHUSER	(GFP_USER | __GFP_HIGHMEM)
350 #define GFP_HIGHUSER_MOVABLE	(GFP_HIGHUSER | __GFP_MOVABLE | \
351 			 __GFP_SKIP_KASAN_POISON)
352 #define GFP_TRANSHUGE_LIGHT	((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
353 			 __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
354 #define GFP_TRANSHUGE	(GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
355 
356 /* Convert GFP flags to their corresponding migrate type */
357 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
358 #define GFP_MOVABLE_SHIFT 3
359 
gfp_migratetype(const gfp_t gfp_flags)360 static inline int gfp_migratetype(const gfp_t gfp_flags)
361 {
362 	VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
363 	BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
364 	BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
365 
366 	if (unlikely(page_group_by_mobility_disabled))
367 		return MIGRATE_UNMOVABLE;
368 
369 	/* Group based on mobility */
370 	return (__force unsigned long)(gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
371 }
372 #undef GFP_MOVABLE_MASK
373 #undef GFP_MOVABLE_SHIFT
374 
gfpflags_allow_blocking(const gfp_t gfp_flags)375 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
376 {
377 	return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
378 }
379 
380 /**
381  * gfpflags_normal_context - is gfp_flags a normal sleepable context?
382  * @gfp_flags: gfp_flags to test
383  *
384  * Test whether @gfp_flags indicates that the allocation is from the
385  * %current context and allowed to sleep.
386  *
387  * An allocation being allowed to block doesn't mean it owns the %current
388  * context.  When direct reclaim path tries to allocate memory, the
389  * allocation context is nested inside whatever %current was doing at the
390  * time of the original allocation.  The nested allocation may be allowed
391  * to block but modifying anything %current owns can corrupt the outer
392  * context's expectations.
393  *
394  * %true result from this function indicates that the allocation context
395  * can sleep and use anything that's associated with %current.
396  */
gfpflags_normal_context(const gfp_t gfp_flags)397 static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
398 {
399 	return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
400 		__GFP_DIRECT_RECLAIM;
401 }
402 
403 #ifdef CONFIG_HIGHMEM
404 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
405 #else
406 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
407 #endif
408 
409 #ifdef CONFIG_ZONE_DMA
410 #define OPT_ZONE_DMA ZONE_DMA
411 #else
412 #define OPT_ZONE_DMA ZONE_NORMAL
413 #endif
414 
415 #ifdef CONFIG_ZONE_DMA32
416 #define OPT_ZONE_DMA32 ZONE_DMA32
417 #else
418 #define OPT_ZONE_DMA32 ZONE_NORMAL
419 #endif
420 
421 /*
422  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
423  * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
424  * bits long and there are 16 of them to cover all possible combinations of
425  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
426  *
427  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
428  * But GFP_MOVABLE is not only a zone specifier but also an allocation
429  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
430  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
431  *
432  *       bit       result
433  *       =================
434  *       0x0    => NORMAL
435  *       0x1    => DMA or NORMAL
436  *       0x2    => HIGHMEM or NORMAL
437  *       0x3    => BAD (DMA+HIGHMEM)
438  *       0x4    => DMA32 or NORMAL
439  *       0x5    => BAD (DMA+DMA32)
440  *       0x6    => BAD (HIGHMEM+DMA32)
441  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
442  *       0x8    => NORMAL (MOVABLE+0)
443  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
444  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
445  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
446  *       0xc    => DMA32 or NORMAL (MOVABLE+DMA32)
447  *       0xd    => BAD (MOVABLE+DMA32+DMA)
448  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
449  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
450  *
451  * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
452  */
453 
454 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
455 /* ZONE_DEVICE is not a valid GFP zone specifier */
456 #define GFP_ZONES_SHIFT 2
457 #else
458 #define GFP_ZONES_SHIFT ZONES_SHIFT
459 #endif
460 
461 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
462 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
463 #endif
464 
465 #define GFP_ZONE_TABLE ( \
466 	(ZONE_NORMAL << 0 * GFP_ZONES_SHIFT)				       \
467 	| (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT)		       \
468 	| (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT)	       \
469 	| (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT)		       \
470 	| (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT)		       \
471 	| (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT)    \
472 	| (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
473 	| (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
474 )
475 
476 /*
477  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
478  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
479  * entry starting with bit 0. Bit is set if the combination is not
480  * allowed.
481  */
482 #define GFP_ZONE_BAD ( \
483 	1 << (___GFP_DMA | ___GFP_HIGHMEM)				      \
484 	| 1 << (___GFP_DMA | ___GFP_DMA32)				      \
485 	| 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)				      \
486 	| 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
487 	| 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)		      \
488 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)		      \
489 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)		      \
490 	| 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
491 )
492 
gfp_zone(gfp_t flags)493 static inline enum zone_type gfp_zone(gfp_t flags)
494 {
495 	enum zone_type z;
496 	int bit = (__force int) (flags & GFP_ZONEMASK);
497 
498 	z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
499 					 ((1 << GFP_ZONES_SHIFT) - 1);
500 	VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
501 	return z;
502 }
503 
504 /*
505  * There is only one page-allocator function, and two main namespaces to
506  * it. The alloc_page*() variants return 'struct page *' and as such
507  * can allocate highmem pages, the *get*page*() variants return
508  * virtual kernel addresses to the allocated page(s).
509  */
510 
gfp_zonelist(gfp_t flags)511 static inline int gfp_zonelist(gfp_t flags)
512 {
513 #ifdef CONFIG_NUMA
514 	if (unlikely(flags & __GFP_THISNODE))
515 		return ZONELIST_NOFALLBACK;
516 #endif
517 	return ZONELIST_FALLBACK;
518 }
519 
520 /*
521  * We get the zone list from the current node and the gfp_mask.
522  * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones.
523  * There are two zonelists per node, one for all zones with memory and
524  * one containing just zones from the node the zonelist belongs to.
525  *
526  * For the case of non-NUMA systems the NODE_DATA() gets optimized to
527  * &contig_page_data at compile-time.
528  */
node_zonelist(int nid,gfp_t flags)529 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
530 {
531 	return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
532 }
533 
534 #ifndef HAVE_ARCH_FREE_PAGE
arch_free_page(struct page * page,int order)535 static inline void arch_free_page(struct page *page, int order) { }
536 #endif
537 #ifndef HAVE_ARCH_ALLOC_PAGE
arch_alloc_page(struct page * page,int order)538 static inline void arch_alloc_page(struct page *page, int order) { }
539 #endif
540 
541 struct page *__alloc_pages(gfp_t gfp, unsigned int order, int preferred_nid,
542 		nodemask_t *nodemask);
543 struct folio *__folio_alloc(gfp_t gfp, unsigned int order, int preferred_nid,
544 		nodemask_t *nodemask);
545 
546 unsigned long __alloc_pages_bulk(gfp_t gfp, int preferred_nid,
547 				nodemask_t *nodemask, int nr_pages,
548 				struct list_head *page_list,
549 				struct page **page_array);
550 
551 unsigned long alloc_pages_bulk_array_mempolicy(gfp_t gfp,
552 				unsigned long nr_pages,
553 				struct page **page_array);
554 
555 /* Bulk allocate order-0 pages */
556 static inline unsigned long
alloc_pages_bulk_list(gfp_t gfp,unsigned long nr_pages,struct list_head * list)557 alloc_pages_bulk_list(gfp_t gfp, unsigned long nr_pages, struct list_head *list)
558 {
559 	return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, list, NULL);
560 }
561 
562 static inline unsigned long
alloc_pages_bulk_array(gfp_t gfp,unsigned long nr_pages,struct page ** page_array)563 alloc_pages_bulk_array(gfp_t gfp, unsigned long nr_pages, struct page **page_array)
564 {
565 	return __alloc_pages_bulk(gfp, numa_mem_id(), NULL, nr_pages, NULL, page_array);
566 }
567 
568 static inline unsigned long
alloc_pages_bulk_array_node(gfp_t gfp,int nid,unsigned long nr_pages,struct page ** page_array)569 alloc_pages_bulk_array_node(gfp_t gfp, int nid, unsigned long nr_pages, struct page **page_array)
570 {
571 	if (nid == NUMA_NO_NODE)
572 		nid = numa_mem_id();
573 
574 	return __alloc_pages_bulk(gfp, nid, NULL, nr_pages, NULL, page_array);
575 }
576 
577 /*
578  * Allocate pages, preferring the node given as nid. The node must be valid and
579  * online. For more general interface, see alloc_pages_node().
580  */
581 static inline struct page *
__alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)582 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
583 {
584 	VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
585 	VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
586 
587 	return __alloc_pages(gfp_mask, order, nid, NULL);
588 }
589 
590 static inline
__folio_alloc_node(gfp_t gfp,unsigned int order,int nid)591 struct folio *__folio_alloc_node(gfp_t gfp, unsigned int order, int nid)
592 {
593 	VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
594 	VM_WARN_ON((gfp & __GFP_THISNODE) && !node_online(nid));
595 
596 	return __folio_alloc(gfp, order, nid, NULL);
597 }
598 
599 /*
600  * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
601  * prefer the current CPU's closest node. Otherwise node must be valid and
602  * online.
603  */
alloc_pages_node(int nid,gfp_t gfp_mask,unsigned int order)604 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
605 						unsigned int order)
606 {
607 	if (nid == NUMA_NO_NODE)
608 		nid = numa_mem_id();
609 
610 	return __alloc_pages_node(nid, gfp_mask, order);
611 }
612 
613 #ifdef CONFIG_NUMA
614 struct page *alloc_pages(gfp_t gfp, unsigned int order);
615 struct folio *folio_alloc(gfp_t gfp, unsigned order);
616 struct folio *vma_alloc_folio(gfp_t gfp, int order, struct vm_area_struct *vma,
617 		unsigned long addr, bool hugepage);
618 #else
alloc_pages(gfp_t gfp_mask,unsigned int order)619 static inline struct page *alloc_pages(gfp_t gfp_mask, unsigned int order)
620 {
621 	return alloc_pages_node(numa_node_id(), gfp_mask, order);
622 }
folio_alloc(gfp_t gfp,unsigned int order)623 static inline struct folio *folio_alloc(gfp_t gfp, unsigned int order)
624 {
625 	return __folio_alloc_node(gfp, order, numa_node_id());
626 }
627 #define vma_alloc_folio(gfp, order, vma, addr, hugepage)		\
628 	folio_alloc(gfp, order)
629 #endif
630 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
alloc_page_vma(gfp_t gfp,struct vm_area_struct * vma,unsigned long addr)631 static inline struct page *alloc_page_vma(gfp_t gfp,
632 		struct vm_area_struct *vma, unsigned long addr)
633 {
634 	struct folio *folio = vma_alloc_folio(gfp, 0, vma, addr, false);
635 
636 	return &folio->page;
637 }
638 
639 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
640 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
641 
642 void *alloc_pages_exact(size_t size, gfp_t gfp_mask) __alloc_size(1);
643 void free_pages_exact(void *virt, size_t size);
644 __meminit void *alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2);
645 
646 #define __get_free_page(gfp_mask) \
647 		__get_free_pages((gfp_mask), 0)
648 
649 #define __get_dma_pages(gfp_mask, order) \
650 		__get_free_pages((gfp_mask) | GFP_DMA, (order))
651 
652 extern void __free_pages(struct page *page, unsigned int order);
653 extern void free_pages(unsigned long addr, unsigned int order);
654 
655 struct page_frag_cache;
656 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
657 extern void *page_frag_alloc_align(struct page_frag_cache *nc,
658 				   unsigned int fragsz, gfp_t gfp_mask,
659 				   unsigned int align_mask);
660 
page_frag_alloc(struct page_frag_cache * nc,unsigned int fragsz,gfp_t gfp_mask)661 static inline void *page_frag_alloc(struct page_frag_cache *nc,
662 			     unsigned int fragsz, gfp_t gfp_mask)
663 {
664 	return page_frag_alloc_align(nc, fragsz, gfp_mask, ~0u);
665 }
666 
667 extern void page_frag_free(void *addr);
668 
669 #define __free_page(page) __free_pages((page), 0)
670 #define free_page(addr) free_pages((addr), 0)
671 
672 void page_alloc_init(void);
673 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
674 void drain_all_pages(struct zone *zone);
675 void drain_local_pages(struct zone *zone);
676 
677 void page_alloc_init_late(void);
678 
679 /*
680  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
681  * GFP flags are used before interrupts are enabled. Once interrupts are
682  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
683  * hibernation, it is used by PM to avoid I/O during memory allocation while
684  * devices are suspended.
685  */
686 extern gfp_t gfp_allowed_mask;
687 
688 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
689 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
690 
691 extern void pm_restrict_gfp_mask(void);
692 extern void pm_restore_gfp_mask(void);
693 
694 extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma);
695 
696 #ifdef CONFIG_PM_SLEEP
697 extern bool pm_suspended_storage(void);
698 #else
pm_suspended_storage(void)699 static inline bool pm_suspended_storage(void)
700 {
701 	return false;
702 }
703 #endif /* CONFIG_PM_SLEEP */
704 
705 #ifdef CONFIG_CONTIG_ALLOC
706 /* The below functions must be run on a range from a single zone. */
707 extern int alloc_contig_range(unsigned long start, unsigned long end,
708 			      unsigned migratetype, gfp_t gfp_mask);
709 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
710 				       int nid, nodemask_t *nodemask);
711 #endif
712 void free_contig_range(unsigned long pfn, unsigned long nr_pages);
713 
714 #ifdef CONFIG_CMA
715 /* CMA stuff */
716 extern void init_cma_reserved_pageblock(struct page *page);
717 #endif
718 
719 #endif /* __LINUX_GFP_H */
720