Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0 */
2#ifndef __LINUX_GFP_H
3#define __LINUX_GFP_H
4
5#include <linux/gfp_types.h>
6
7#include <linux/mmzone.h>
8#include <linux/topology.h>
9#include <linux/alloc_tag.h>
10#include <linux/cleanup.h>
11#include <linux/sched.h>
12
13struct vm_area_struct;
14struct mempolicy;
15
16/* Helper macro to avoid gfp flags if they are the default one */
17#define __default_gfp(a,b,...) b
18#define default_gfp(...) __default_gfp(,##__VA_ARGS__,GFP_KERNEL)
19
20/* Convert GFP flags to their corresponding migrate type */
21#define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
22#define GFP_MOVABLE_SHIFT 3
23
24static inline int gfp_migratetype(const gfp_t gfp_flags)
25{
26 VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
27 BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
28 BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
29 BUILD_BUG_ON((___GFP_RECLAIMABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_RECLAIMABLE);
30 BUILD_BUG_ON(((___GFP_MOVABLE | ___GFP_RECLAIMABLE) >>
31 GFP_MOVABLE_SHIFT) != MIGRATE_HIGHATOMIC);
32
33 if (unlikely(page_group_by_mobility_disabled))
34 return MIGRATE_UNMOVABLE;
35
36 /* Group based on mobility */
37 return (__force unsigned long)(gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
38}
39#undef GFP_MOVABLE_MASK
40#undef GFP_MOVABLE_SHIFT
41
42static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
43{
44 return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
45}
46
47static inline bool gfpflags_allow_spinning(const gfp_t gfp_flags)
48{
49 /*
50 * !__GFP_DIRECT_RECLAIM -> direct claim is not allowed.
51 * !__GFP_KSWAPD_RECLAIM -> it's not safe to wake up kswapd.
52 * All GFP_* flags including GFP_NOWAIT use one or both flags.
53 * alloc_pages_nolock() is the only API that doesn't specify either flag.
54 *
55 * This is stronger than GFP_NOWAIT or GFP_ATOMIC because
56 * those are guaranteed to never block on a sleeping lock.
57 * Here we are enforcing that the allocation doesn't ever spin
58 * on any locks (i.e. only trylocks). There is no high level
59 * GFP_$FOO flag for this use in alloc_pages_nolock() as the
60 * regular page allocator doesn't fully support this
61 * allocation mode.
62 */
63 return !!(gfp_flags & __GFP_RECLAIM);
64}
65
66#ifdef CONFIG_HIGHMEM
67#define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
68#else
69#define OPT_ZONE_HIGHMEM ZONE_NORMAL
70#endif
71
72#ifdef CONFIG_ZONE_DMA
73#define OPT_ZONE_DMA ZONE_DMA
74#else
75#define OPT_ZONE_DMA ZONE_NORMAL
76#endif
77
78#ifdef CONFIG_ZONE_DMA32
79#define OPT_ZONE_DMA32 ZONE_DMA32
80#else
81#define OPT_ZONE_DMA32 ZONE_NORMAL
82#endif
83
84/*
85 * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
86 * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
87 * bits long and there are 16 of them to cover all possible combinations of
88 * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
89 *
90 * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
91 * But GFP_MOVABLE is not only a zone specifier but also an allocation
92 * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
93 * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
94 *
95 * bit result
96 * =================
97 * 0x0 => NORMAL
98 * 0x1 => DMA or NORMAL
99 * 0x2 => HIGHMEM or NORMAL
100 * 0x3 => BAD (DMA+HIGHMEM)
101 * 0x4 => DMA32 or NORMAL
102 * 0x5 => BAD (DMA+DMA32)
103 * 0x6 => BAD (HIGHMEM+DMA32)
104 * 0x7 => BAD (HIGHMEM+DMA32+DMA)
105 * 0x8 => NORMAL (MOVABLE+0)
106 * 0x9 => DMA or NORMAL (MOVABLE+DMA)
107 * 0xa => MOVABLE (Movable is valid only if HIGHMEM is set too)
108 * 0xb => BAD (MOVABLE+HIGHMEM+DMA)
109 * 0xc => DMA32 or NORMAL (MOVABLE+DMA32)
110 * 0xd => BAD (MOVABLE+DMA32+DMA)
111 * 0xe => BAD (MOVABLE+DMA32+HIGHMEM)
112 * 0xf => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
113 *
114 * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
115 */
116
117#if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
118/* ZONE_DEVICE is not a valid GFP zone specifier */
119#define GFP_ZONES_SHIFT 2
120#else
121#define GFP_ZONES_SHIFT ZONES_SHIFT
122#endif
123
124#if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
125#error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
126#endif
127
128#define GFP_ZONE_TABLE ( \
129 (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT) \
130 | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT) \
131 | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT) \
132 | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT) \
133 | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT) \
134 | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT) \
135 | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
136 | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
137)
138
139/*
140 * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
141 * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
142 * entry starting with bit 0. Bit is set if the combination is not
143 * allowed.
144 */
145#define GFP_ZONE_BAD ( \
146 1 << (___GFP_DMA | ___GFP_HIGHMEM) \
147 | 1 << (___GFP_DMA | ___GFP_DMA32) \
148 | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM) \
149 | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM) \
150 | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA) \
151 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA) \
152 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM) \
153 | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM) \
154)
155
156static inline enum zone_type gfp_zone(gfp_t flags)
157{
158 enum zone_type z;
159 int bit = (__force int) (flags & GFP_ZONEMASK);
160
161 z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
162 ((1 << GFP_ZONES_SHIFT) - 1);
163 VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
164 return z;
165}
166
167/*
168 * There is only one page-allocator function, and two main namespaces to
169 * it. The alloc_page*() variants return 'struct page *' and as such
170 * can allocate highmem pages, the *get*page*() variants return
171 * virtual kernel addresses to the allocated page(s).
172 */
173
174static inline int gfp_zonelist(gfp_t flags)
175{
176#ifdef CONFIG_NUMA
177 if (unlikely(flags & __GFP_THISNODE))
178 return ZONELIST_NOFALLBACK;
179#endif
180 return ZONELIST_FALLBACK;
181}
182
183/*
184 * gfp flag masking for nested internal allocations.
185 *
186 * For code that needs to do allocations inside the public allocation API (e.g.
187 * memory allocation tracking code) the allocations need to obey the caller
188 * allocation context constrains to prevent allocation context mismatches (e.g.
189 * GFP_KERNEL allocations in GFP_NOFS contexts) from potential deadlock
190 * situations.
191 *
192 * It is also assumed that these nested allocations are for internal kernel
193 * object storage purposes only and are not going to be used for DMA, etc. Hence
194 * we strip out all the zone information and leave just the context information
195 * intact.
196 *
197 * Further, internal allocations must fail before the higher level allocation
198 * can fail, so we must make them fail faster and fail silently. We also don't
199 * want them to deplete emergency reserves. Hence nested allocations must be
200 * prepared for these allocations to fail.
201 */
202static inline gfp_t gfp_nested_mask(gfp_t flags)
203{
204 return ((flags & (GFP_KERNEL | GFP_ATOMIC | __GFP_NOLOCKDEP)) |
205 (__GFP_NORETRY | __GFP_NOMEMALLOC | __GFP_NOWARN));
206}
207
208/*
209 * We get the zone list from the current node and the gfp_mask.
210 * This zone list contains a maximum of MAX_NUMNODES*MAX_NR_ZONES zones.
211 * There are two zonelists per node, one for all zones with memory and
212 * one containing just zones from the node the zonelist belongs to.
213 *
214 * For the case of non-NUMA systems the NODE_DATA() gets optimized to
215 * &contig_page_data at compile-time.
216 */
217static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
218{
219 return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
220}
221
222#ifndef HAVE_ARCH_FREE_PAGE
223static inline void arch_free_page(struct page *page, int order) { }
224#endif
225#ifndef HAVE_ARCH_ALLOC_PAGE
226static inline void arch_alloc_page(struct page *page, int order) { }
227#endif
228
229struct page *__alloc_pages_noprof(gfp_t gfp, unsigned int order, int preferred_nid,
230 nodemask_t *nodemask);
231#define __alloc_pages(...) alloc_hooks(__alloc_pages_noprof(__VA_ARGS__))
232
233struct folio *__folio_alloc_noprof(gfp_t gfp, unsigned int order, int preferred_nid,
234 nodemask_t *nodemask);
235#define __folio_alloc(...) alloc_hooks(__folio_alloc_noprof(__VA_ARGS__))
236
237unsigned long alloc_pages_bulk_noprof(gfp_t gfp, int preferred_nid,
238 nodemask_t *nodemask, int nr_pages,
239 struct page **page_array);
240#define __alloc_pages_bulk(...) alloc_hooks(alloc_pages_bulk_noprof(__VA_ARGS__))
241
242unsigned long alloc_pages_bulk_mempolicy_noprof(gfp_t gfp,
243 unsigned long nr_pages,
244 struct page **page_array);
245#define alloc_pages_bulk_mempolicy(...) \
246 alloc_hooks(alloc_pages_bulk_mempolicy_noprof(__VA_ARGS__))
247
248/* Bulk allocate order-0 pages */
249#define alloc_pages_bulk(_gfp, _nr_pages, _page_array) \
250 __alloc_pages_bulk(_gfp, numa_mem_id(), NULL, _nr_pages, _page_array)
251
252static inline unsigned long
253alloc_pages_bulk_node_noprof(gfp_t gfp, int nid, unsigned long nr_pages,
254 struct page **page_array)
255{
256 if (nid == NUMA_NO_NODE)
257 nid = numa_mem_id();
258
259 return alloc_pages_bulk_noprof(gfp, nid, NULL, nr_pages, page_array);
260}
261
262#define alloc_pages_bulk_node(...) \
263 alloc_hooks(alloc_pages_bulk_node_noprof(__VA_ARGS__))
264
265static inline void warn_if_node_offline(int this_node, gfp_t gfp_mask)
266{
267 gfp_t warn_gfp = gfp_mask & (__GFP_THISNODE|__GFP_NOWARN);
268
269 if (warn_gfp != (__GFP_THISNODE|__GFP_NOWARN))
270 return;
271
272 if (node_online(this_node))
273 return;
274
275 pr_warn("%pGg allocation from offline node %d\n", &gfp_mask, this_node);
276 dump_stack();
277}
278
279/*
280 * Allocate pages, preferring the node given as nid. The node must be valid and
281 * online. For more general interface, see alloc_pages_node().
282 */
283static inline struct page *
284__alloc_pages_node_noprof(int nid, gfp_t gfp_mask, unsigned int order)
285{
286 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
287 warn_if_node_offline(nid, gfp_mask);
288
289 return __alloc_pages_noprof(gfp_mask, order, nid, NULL);
290}
291
292#define __alloc_pages_node(...) alloc_hooks(__alloc_pages_node_noprof(__VA_ARGS__))
293
294static inline
295struct folio *__folio_alloc_node_noprof(gfp_t gfp, unsigned int order, int nid)
296{
297 VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
298 warn_if_node_offline(nid, gfp);
299
300 return __folio_alloc_noprof(gfp, order, nid, NULL);
301}
302
303#define __folio_alloc_node(...) alloc_hooks(__folio_alloc_node_noprof(__VA_ARGS__))
304
305/*
306 * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
307 * prefer the current CPU's closest node. Otherwise node must be valid and
308 * online.
309 */
310static inline struct page *alloc_pages_node_noprof(int nid, gfp_t gfp_mask,
311 unsigned int order)
312{
313 if (nid == NUMA_NO_NODE)
314 nid = numa_mem_id();
315
316 return __alloc_pages_node_noprof(nid, gfp_mask, order);
317}
318
319#define alloc_pages_node(...) alloc_hooks(alloc_pages_node_noprof(__VA_ARGS__))
320
321#ifdef CONFIG_NUMA
322struct page *alloc_pages_noprof(gfp_t gfp, unsigned int order);
323struct folio *folio_alloc_noprof(gfp_t gfp, unsigned int order);
324struct folio *folio_alloc_mpol_noprof(gfp_t gfp, unsigned int order,
325 struct mempolicy *mpol, pgoff_t ilx, int nid);
326struct folio *vma_alloc_folio_noprof(gfp_t gfp, int order, struct vm_area_struct *vma,
327 unsigned long addr);
328#else
329static inline struct page *alloc_pages_noprof(gfp_t gfp_mask, unsigned int order)
330{
331 return alloc_pages_node_noprof(numa_node_id(), gfp_mask, order);
332}
333static inline struct folio *folio_alloc_noprof(gfp_t gfp, unsigned int order)
334{
335 return __folio_alloc_node_noprof(gfp, order, numa_node_id());
336}
337static inline struct folio *folio_alloc_mpol_noprof(gfp_t gfp, unsigned int order,
338 struct mempolicy *mpol, pgoff_t ilx, int nid)
339{
340 return folio_alloc_noprof(gfp, order);
341}
342static inline struct folio *vma_alloc_folio_noprof(gfp_t gfp, int order,
343 struct vm_area_struct *vma, unsigned long addr)
344{
345 return folio_alloc_noprof(gfp, order);
346}
347#endif
348
349#define alloc_pages(...) alloc_hooks(alloc_pages_noprof(__VA_ARGS__))
350#define folio_alloc(...) alloc_hooks(folio_alloc_noprof(__VA_ARGS__))
351#define folio_alloc_mpol(...) alloc_hooks(folio_alloc_mpol_noprof(__VA_ARGS__))
352#define vma_alloc_folio(...) alloc_hooks(vma_alloc_folio_noprof(__VA_ARGS__))
353
354#define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
355
356static inline struct page *alloc_page_vma_noprof(gfp_t gfp,
357 struct vm_area_struct *vma, unsigned long addr)
358{
359 struct folio *folio = vma_alloc_folio_noprof(gfp, 0, vma, addr);
360
361 return &folio->page;
362}
363#define alloc_page_vma(...) alloc_hooks(alloc_page_vma_noprof(__VA_ARGS__))
364
365struct page *alloc_pages_nolock_noprof(gfp_t gfp_flags, int nid, unsigned int order);
366#define alloc_pages_nolock(...) alloc_hooks(alloc_pages_nolock_noprof(__VA_ARGS__))
367
368extern unsigned long get_free_pages_noprof(gfp_t gfp_mask, unsigned int order);
369#define __get_free_pages(...) alloc_hooks(get_free_pages_noprof(__VA_ARGS__))
370
371extern unsigned long get_zeroed_page_noprof(gfp_t gfp_mask);
372#define get_zeroed_page(...) alloc_hooks(get_zeroed_page_noprof(__VA_ARGS__))
373
374void *alloc_pages_exact_noprof(size_t size, gfp_t gfp_mask) __alloc_size(1);
375#define alloc_pages_exact(...) alloc_hooks(alloc_pages_exact_noprof(__VA_ARGS__))
376
377void free_pages_exact(void *virt, size_t size);
378
379__meminit void *alloc_pages_exact_nid_noprof(int nid, size_t size, gfp_t gfp_mask) __alloc_size(2);
380#define alloc_pages_exact_nid(...) \
381 alloc_hooks(alloc_pages_exact_nid_noprof(__VA_ARGS__))
382
383#define __get_free_page(gfp_mask) \
384 __get_free_pages((gfp_mask), 0)
385
386#define __get_dma_pages(gfp_mask, order) \
387 __get_free_pages((gfp_mask) | GFP_DMA, (order))
388
389extern void __free_pages(struct page *page, unsigned int order);
390extern void free_pages_nolock(struct page *page, unsigned int order);
391extern void free_pages(unsigned long addr, unsigned int order);
392
393#define __free_page(page) __free_pages((page), 0)
394#define free_page(addr) free_pages((addr), 0)
395
396void page_alloc_init_cpuhp(void);
397bool decay_pcp_high(struct zone *zone, struct per_cpu_pages *pcp);
398void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
399void drain_all_pages(struct zone *zone);
400void drain_local_pages(struct zone *zone);
401
402void page_alloc_init_late(void);
403void setup_pcp_cacheinfo(unsigned int cpu);
404
405/*
406 * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
407 * GFP flags are used before interrupts are enabled. Once interrupts are
408 * enabled, it is set to __GFP_BITS_MASK while the system is running. During
409 * hibernation, it is used by PM to avoid I/O during memory allocation while
410 * devices are suspended.
411 */
412extern gfp_t gfp_allowed_mask;
413
414/* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
415bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
416
417/* A helper for checking if gfp includes all the specified flags */
418static inline bool gfp_has_flags(gfp_t gfp, gfp_t flags)
419{
420 return (gfp & flags) == flags;
421}
422
423static inline bool gfp_has_io_fs(gfp_t gfp)
424{
425 return gfp_has_flags(gfp, __GFP_IO | __GFP_FS);
426}
427
428/*
429 * Check if the gfp flags allow compaction - GFP_NOIO is a really
430 * tricky context because the migration might require IO.
431 */
432static inline bool gfp_compaction_allowed(gfp_t gfp_mask)
433{
434 return IS_ENABLED(CONFIG_COMPACTION) && (gfp_mask & __GFP_IO);
435}
436
437extern gfp_t vma_thp_gfp_mask(struct vm_area_struct *vma);
438
439#ifdef CONFIG_CONTIG_ALLOC
440
441typedef unsigned int __bitwise acr_flags_t;
442#define ACR_FLAGS_NONE ((__force acr_flags_t)0) // ordinary allocation request
443#define ACR_FLAGS_CMA ((__force acr_flags_t)BIT(0)) // allocate for CMA
444
445/* The below functions must be run on a range from a single zone. */
446int alloc_contig_frozen_range_noprof(unsigned long start, unsigned long end,
447 acr_flags_t alloc_flags, gfp_t gfp_mask);
448#define alloc_contig_frozen_range(...) \
449 alloc_hooks(alloc_contig_frozen_range_noprof(__VA_ARGS__))
450
451int alloc_contig_range_noprof(unsigned long start, unsigned long end,
452 acr_flags_t alloc_flags, gfp_t gfp_mask);
453#define alloc_contig_range(...) \
454 alloc_hooks(alloc_contig_range_noprof(__VA_ARGS__))
455
456struct page *alloc_contig_frozen_pages_noprof(unsigned long nr_pages,
457 gfp_t gfp_mask, int nid, nodemask_t *nodemask);
458#define alloc_contig_frozen_pages(...) \
459 alloc_hooks(alloc_contig_frozen_pages_noprof(__VA_ARGS__))
460
461struct page *alloc_contig_pages_noprof(unsigned long nr_pages, gfp_t gfp_mask,
462 int nid, nodemask_t *nodemask);
463#define alloc_contig_pages(...) \
464 alloc_hooks(alloc_contig_pages_noprof(__VA_ARGS__))
465
466void free_contig_frozen_range(unsigned long pfn, unsigned long nr_pages);
467void free_contig_range(unsigned long pfn, unsigned long nr_pages);
468#endif
469
470DEFINE_FREE(free_page, void *, free_page((unsigned long)_T))
471
472#endif /* __LINUX_GFP_H */