~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/include/linux/gfp.h

Version: ~ [ linux-5.13-rc1 ] ~ [ linux-5.12.2 ] ~ [ linux-5.11.19 ] ~ [ linux-5.10.35 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.117 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.190 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.232 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.268 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.268 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.18.140 ] ~ [ linux-3.16.85 ] ~ [ linux-3.14.79 ] ~ [ linux-3.12.74 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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_ZERO             0x100u
 28 #define ___GFP_ATOMIC           0x200u
 29 #define ___GFP_DIRECT_RECLAIM   0x400u
 30 #define ___GFP_KSWAPD_RECLAIM   0x800u
 31 #define ___GFP_WRITE            0x1000u
 32 #define ___GFP_NOWARN           0x2000u
 33 #define ___GFP_RETRY_MAYFAIL    0x4000u
 34 #define ___GFP_NOFAIL           0x8000u
 35 #define ___GFP_NORETRY          0x10000u
 36 #define ___GFP_MEMALLOC         0x20000u
 37 #define ___GFP_COMP             0x40000u
 38 #define ___GFP_NOMEMALLOC       0x80000u
 39 #define ___GFP_HARDWALL         0x100000u
 40 #define ___GFP_THISNODE         0x200000u
 41 #define ___GFP_ACCOUNT          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 satisfied 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  * Please note that all the following flags are only applicable to sleepable
128  * allocations (e.g. %GFP_NOWAIT and %GFP_ATOMIC will ignore them).
129  *
130  * %__GFP_IO can start physical IO.
131  *
132  * %__GFP_FS can call down to the low-level FS. Clearing the flag avoids the
133  * allocator recursing into the filesystem which might already be holding
134  * locks.
135  *
136  * %__GFP_DIRECT_RECLAIM indicates that the caller may enter direct reclaim.
137  * This flag can be cleared to avoid unnecessary delays when a fallback
138  * option is available.
139  *
140  * %__GFP_KSWAPD_RECLAIM indicates that the caller wants to wake kswapd when
141  * the low watermark is reached and have it reclaim pages until the high
142  * watermark is reached. A caller may wish to clear this flag when fallback
143  * options are available and the reclaim is likely to disrupt the system. The
144  * canonical example is THP allocation where a fallback is cheap but
145  * reclaim/compaction may cause indirect stalls.
146  *
147  * %__GFP_RECLAIM is shorthand to allow/forbid both direct and kswapd reclaim.
148  *
149  * The default allocator behavior depends on the request size. We have a concept
150  * of so called costly allocations (with order > %PAGE_ALLOC_COSTLY_ORDER).
151  * !costly allocations are too essential to fail so they are implicitly
152  * non-failing by default (with some exceptions like OOM victims might fail so
153  * the caller still has to check for failures) while costly requests try to be
154  * not disruptive and back off even without invoking the OOM killer.
155  * The following three modifiers might be used to override some of these
156  * implicit rules
157  *
158  * %__GFP_NORETRY: The VM implementation will try only very lightweight
159  * memory direct reclaim to get some memory under memory pressure (thus
160  * it can sleep). It will avoid disruptive actions like OOM killer. The
161  * caller must handle the failure which is quite likely to happen under
162  * heavy memory pressure. The flag is suitable when failure can easily be
163  * handled at small cost, such as reduced throughput
164  *
165  * %__GFP_RETRY_MAYFAIL: The VM implementation will retry memory reclaim
166  * procedures that have previously failed if there is some indication
167  * that progress has been made else where.  It can wait for other
168  * tasks to attempt high level approaches to freeing memory such as
169  * compaction (which removes fragmentation) and page-out.
170  * There is still a definite limit to the number of retries, but it is
171  * a larger limit than with %__GFP_NORETRY.
172  * Allocations with this flag may fail, but only when there is
173  * genuinely little unused memory. While these allocations do not
174  * directly trigger the OOM killer, their failure indicates that
175  * the system is likely to need to use the OOM killer soon.  The
176  * caller must handle failure, but can reasonably do so by failing
177  * a higher-level request, or completing it only in a much less
178  * efficient manner.
179  * If the allocation does fail, and the caller is in a position to
180  * free some non-essential memory, doing so could benefit the system
181  * as a whole.
182  *
183  * %__GFP_NOFAIL: The VM implementation _must_ retry infinitely: the caller
184  * cannot handle allocation failures. The allocation could block
185  * indefinitely but will never return with failure. Testing for
186  * failure is pointless.
187  * New users should be evaluated carefully (and the flag should be
188  * used only when there is no reasonable failure policy) but it is
189  * definitely preferable to use the flag rather than opencode endless
190  * loop around allocator.
191  * Using this flag for costly allocations is _highly_ discouraged.
192  */
193 #define __GFP_IO        ((__force gfp_t)___GFP_IO)
194 #define __GFP_FS        ((__force gfp_t)___GFP_FS)
195 #define __GFP_DIRECT_RECLAIM    ((__force gfp_t)___GFP_DIRECT_RECLAIM) /* Caller can reclaim */
196 #define __GFP_KSWAPD_RECLAIM    ((__force gfp_t)___GFP_KSWAPD_RECLAIM) /* kswapd can wake */
197 #define __GFP_RECLAIM ((__force gfp_t)(___GFP_DIRECT_RECLAIM|___GFP_KSWAPD_RECLAIM))
198 #define __GFP_RETRY_MAYFAIL     ((__force gfp_t)___GFP_RETRY_MAYFAIL)
199 #define __GFP_NOFAIL    ((__force gfp_t)___GFP_NOFAIL)
200 #define __GFP_NORETRY   ((__force gfp_t)___GFP_NORETRY)
201 
202 /**
203  * DOC: Action modifiers
204  *
205  * Action modifiers
206  * ~~~~~~~~~~~~~~~~
207  *
208  * %__GFP_NOWARN suppresses allocation failure reports.
209  *
210  * %__GFP_COMP address compound page metadata.
211  *
212  * %__GFP_ZERO returns a zeroed page on success.
213  */
214 #define __GFP_NOWARN    ((__force gfp_t)___GFP_NOWARN)
215 #define __GFP_COMP      ((__force gfp_t)___GFP_COMP)
216 #define __GFP_ZERO      ((__force gfp_t)___GFP_ZERO)
217 
218 /* Disable lockdep for GFP context tracking */
219 #define __GFP_NOLOCKDEP ((__force gfp_t)___GFP_NOLOCKDEP)
220 
221 /* Room for N __GFP_FOO bits */
222 #define __GFP_BITS_SHIFT (23 + IS_ENABLED(CONFIG_LOCKDEP))
223 #define __GFP_BITS_MASK ((__force gfp_t)((1 << __GFP_BITS_SHIFT) - 1))
224 
225 /**
226  * DOC: Useful GFP flag combinations
227  *
228  * Useful GFP flag combinations
229  * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~
230  *
231  * Useful GFP flag combinations that are commonly used. It is recommended
232  * that subsystems start with one of these combinations and then set/clear
233  * %__GFP_FOO flags as necessary.
234  *
235  * %GFP_ATOMIC users can not sleep and need the allocation to succeed. A lower
236  * watermark is applied to allow access to "atomic reserves"
237  *
238  * %GFP_KERNEL is typical for kernel-internal allocations. The caller requires
239  * %ZONE_NORMAL or a lower zone for direct access but can direct reclaim.
240  *
241  * %GFP_KERNEL_ACCOUNT is the same as GFP_KERNEL, except the allocation is
242  * accounted to kmemcg.
243  *
244  * %GFP_NOWAIT is for kernel allocations that should not stall for direct
245  * reclaim, start physical IO or use any filesystem callback.
246  *
247  * %GFP_NOIO will use direct reclaim to discard clean pages or slab pages
248  * that do not require the starting of any physical IO.
249  * Please try to avoid using this flag directly and instead use
250  * memalloc_noio_{save,restore} to mark the whole scope which cannot
251  * perform any IO with a short explanation why. All allocation requests
252  * will inherit GFP_NOIO implicitly.
253  *
254  * %GFP_NOFS will use direct reclaim but will not use any filesystem interfaces.
255  * Please try to avoid using this flag directly and instead use
256  * memalloc_nofs_{save,restore} to mark the whole scope which cannot/shouldn't
257  * recurse into the FS layer with a short explanation why. All allocation
258  * requests will inherit GFP_NOFS implicitly.
259  *
260  * %GFP_USER is for userspace allocations that also need to be directly
261  * accessibly by the kernel or hardware. It is typically used by hardware
262  * for buffers that are mapped to userspace (e.g. graphics) that hardware
263  * still must DMA to. cpuset limits are enforced for these allocations.
264  *
265  * %GFP_DMA exists for historical reasons and should be avoided where possible.
266  * The flags indicates that the caller requires that the lowest zone be
267  * used (%ZONE_DMA or 16M on x86-64). Ideally, this would be removed but
268  * it would require careful auditing as some users really require it and
269  * others use the flag to avoid lowmem reserves in %ZONE_DMA and treat the
270  * lowest zone as a type of emergency reserve.
271  *
272  * %GFP_DMA32 is similar to %GFP_DMA except that the caller requires a 32-bit
273  * address.
274  *
275  * %GFP_HIGHUSER is for userspace allocations that may be mapped to userspace,
276  * do not need to be directly accessible by the kernel but that cannot
277  * move once in use. An example may be a hardware allocation that maps
278  * data directly into userspace but has no addressing limitations.
279  *
280  * %GFP_HIGHUSER_MOVABLE is for userspace allocations that the kernel does not
281  * need direct access to but can use kmap() when access is required. They
282  * are expected to be movable via page reclaim or page migration. Typically,
283  * pages on the LRU would also be allocated with %GFP_HIGHUSER_MOVABLE.
284  *
285  * %GFP_TRANSHUGE and %GFP_TRANSHUGE_LIGHT are used for THP allocations. They
286  * are compound allocations that will generally fail quickly if memory is not
287  * available and will not wake kswapd/kcompactd on failure. The _LIGHT
288  * version does not attempt reclaim/compaction at all and is by default used
289  * in page fault path, while the non-light is used by khugepaged.
290  */
291 #define GFP_ATOMIC      (__GFP_HIGH|__GFP_ATOMIC|__GFP_KSWAPD_RECLAIM)
292 #define GFP_KERNEL      (__GFP_RECLAIM | __GFP_IO | __GFP_FS)
293 #define GFP_KERNEL_ACCOUNT (GFP_KERNEL | __GFP_ACCOUNT)
294 #define GFP_NOWAIT      (__GFP_KSWAPD_RECLAIM)
295 #define GFP_NOIO        (__GFP_RECLAIM)
296 #define GFP_NOFS        (__GFP_RECLAIM | __GFP_IO)
297 #define GFP_USER        (__GFP_RECLAIM | __GFP_IO | __GFP_FS | __GFP_HARDWALL)
298 #define GFP_DMA         __GFP_DMA
299 #define GFP_DMA32       __GFP_DMA32
300 #define GFP_HIGHUSER    (GFP_USER | __GFP_HIGHMEM)
301 #define GFP_HIGHUSER_MOVABLE    (GFP_HIGHUSER | __GFP_MOVABLE)
302 #define GFP_TRANSHUGE_LIGHT     ((GFP_HIGHUSER_MOVABLE | __GFP_COMP | \
303                          __GFP_NOMEMALLOC | __GFP_NOWARN) & ~__GFP_RECLAIM)
304 #define GFP_TRANSHUGE   (GFP_TRANSHUGE_LIGHT | __GFP_DIRECT_RECLAIM)
305 
306 /* Convert GFP flags to their corresponding migrate type */
307 #define GFP_MOVABLE_MASK (__GFP_RECLAIMABLE|__GFP_MOVABLE)
308 #define GFP_MOVABLE_SHIFT 3
309 
310 static inline int gfpflags_to_migratetype(const gfp_t gfp_flags)
311 {
312         VM_WARN_ON((gfp_flags & GFP_MOVABLE_MASK) == GFP_MOVABLE_MASK);
313         BUILD_BUG_ON((1UL << GFP_MOVABLE_SHIFT) != ___GFP_MOVABLE);
314         BUILD_BUG_ON((___GFP_MOVABLE >> GFP_MOVABLE_SHIFT) != MIGRATE_MOVABLE);
315 
316         if (unlikely(page_group_by_mobility_disabled))
317                 return MIGRATE_UNMOVABLE;
318 
319         /* Group based on mobility */
320         return (gfp_flags & GFP_MOVABLE_MASK) >> GFP_MOVABLE_SHIFT;
321 }
322 #undef GFP_MOVABLE_MASK
323 #undef GFP_MOVABLE_SHIFT
324 
325 static inline bool gfpflags_allow_blocking(const gfp_t gfp_flags)
326 {
327         return !!(gfp_flags & __GFP_DIRECT_RECLAIM);
328 }
329 
330 /**
331  * gfpflags_normal_context - is gfp_flags a normal sleepable context?
332  * @gfp_flags: gfp_flags to test
333  *
334  * Test whether @gfp_flags indicates that the allocation is from the
335  * %current context and allowed to sleep.
336  *
337  * An allocation being allowed to block doesn't mean it owns the %current
338  * context.  When direct reclaim path tries to allocate memory, the
339  * allocation context is nested inside whatever %current was doing at the
340  * time of the original allocation.  The nested allocation may be allowed
341  * to block but modifying anything %current owns can corrupt the outer
342  * context's expectations.
343  *
344  * %true result from this function indicates that the allocation context
345  * can sleep and use anything that's associated with %current.
346  */
347 static inline bool gfpflags_normal_context(const gfp_t gfp_flags)
348 {
349         return (gfp_flags & (__GFP_DIRECT_RECLAIM | __GFP_MEMALLOC)) ==
350                 __GFP_DIRECT_RECLAIM;
351 }
352 
353 #ifdef CONFIG_HIGHMEM
354 #define OPT_ZONE_HIGHMEM ZONE_HIGHMEM
355 #else
356 #define OPT_ZONE_HIGHMEM ZONE_NORMAL
357 #endif
358 
359 #ifdef CONFIG_ZONE_DMA
360 #define OPT_ZONE_DMA ZONE_DMA
361 #else
362 #define OPT_ZONE_DMA ZONE_NORMAL
363 #endif
364 
365 #ifdef CONFIG_ZONE_DMA32
366 #define OPT_ZONE_DMA32 ZONE_DMA32
367 #else
368 #define OPT_ZONE_DMA32 ZONE_NORMAL
369 #endif
370 
371 /*
372  * GFP_ZONE_TABLE is a word size bitstring that is used for looking up the
373  * zone to use given the lowest 4 bits of gfp_t. Entries are GFP_ZONES_SHIFT
374  * bits long and there are 16 of them to cover all possible combinations of
375  * __GFP_DMA, __GFP_DMA32, __GFP_MOVABLE and __GFP_HIGHMEM.
376  *
377  * The zone fallback order is MOVABLE=>HIGHMEM=>NORMAL=>DMA32=>DMA.
378  * But GFP_MOVABLE is not only a zone specifier but also an allocation
379  * policy. Therefore __GFP_MOVABLE plus another zone selector is valid.
380  * Only 1 bit of the lowest 3 bits (DMA,DMA32,HIGHMEM) can be set to "1".
381  *
382  *       bit       result
383  *       =================
384  *       0x0    => NORMAL
385  *       0x1    => DMA or NORMAL
386  *       0x2    => HIGHMEM or NORMAL
387  *       0x3    => BAD (DMA+HIGHMEM)
388  *       0x4    => DMA32 or NORMAL
389  *       0x5    => BAD (DMA+DMA32)
390  *       0x6    => BAD (HIGHMEM+DMA32)
391  *       0x7    => BAD (HIGHMEM+DMA32+DMA)
392  *       0x8    => NORMAL (MOVABLE+0)
393  *       0x9    => DMA or NORMAL (MOVABLE+DMA)
394  *       0xa    => MOVABLE (Movable is valid only if HIGHMEM is set too)
395  *       0xb    => BAD (MOVABLE+HIGHMEM+DMA)
396  *       0xc    => DMA32 or NORMAL (MOVABLE+DMA32)
397  *       0xd    => BAD (MOVABLE+DMA32+DMA)
398  *       0xe    => BAD (MOVABLE+DMA32+HIGHMEM)
399  *       0xf    => BAD (MOVABLE+DMA32+HIGHMEM+DMA)
400  *
401  * GFP_ZONES_SHIFT must be <= 2 on 32 bit platforms.
402  */
403 
404 #if defined(CONFIG_ZONE_DEVICE) && (MAX_NR_ZONES-1) <= 4
405 /* ZONE_DEVICE is not a valid GFP zone specifier */
406 #define GFP_ZONES_SHIFT 2
407 #else
408 #define GFP_ZONES_SHIFT ZONES_SHIFT
409 #endif
410 
411 #if 16 * GFP_ZONES_SHIFT > BITS_PER_LONG
412 #error GFP_ZONES_SHIFT too large to create GFP_ZONE_TABLE integer
413 #endif
414 
415 #define GFP_ZONE_TABLE ( \
416         (ZONE_NORMAL << 0 * GFP_ZONES_SHIFT)                                   \
417         | (OPT_ZONE_DMA << ___GFP_DMA * GFP_ZONES_SHIFT)                       \
418         | (OPT_ZONE_HIGHMEM << ___GFP_HIGHMEM * GFP_ZONES_SHIFT)               \
419         | (OPT_ZONE_DMA32 << ___GFP_DMA32 * GFP_ZONES_SHIFT)                   \
420         | (ZONE_NORMAL << ___GFP_MOVABLE * GFP_ZONES_SHIFT)                    \
421         | (OPT_ZONE_DMA << (___GFP_MOVABLE | ___GFP_DMA) * GFP_ZONES_SHIFT)    \
422         | (ZONE_MOVABLE << (___GFP_MOVABLE | ___GFP_HIGHMEM) * GFP_ZONES_SHIFT)\
423         | (OPT_ZONE_DMA32 << (___GFP_MOVABLE | ___GFP_DMA32) * GFP_ZONES_SHIFT)\
424 )
425 
426 /*
427  * GFP_ZONE_BAD is a bitmap for all combinations of __GFP_DMA, __GFP_DMA32
428  * __GFP_HIGHMEM and __GFP_MOVABLE that are not permitted. One flag per
429  * entry starting with bit 0. Bit is set if the combination is not
430  * allowed.
431  */
432 #define GFP_ZONE_BAD ( \
433         1 << (___GFP_DMA | ___GFP_HIGHMEM)                                    \
434         | 1 << (___GFP_DMA | ___GFP_DMA32)                                    \
435         | 1 << (___GFP_DMA32 | ___GFP_HIGHMEM)                                \
436         | 1 << (___GFP_DMA | ___GFP_DMA32 | ___GFP_HIGHMEM)                   \
437         | 1 << (___GFP_MOVABLE | ___GFP_HIGHMEM | ___GFP_DMA)                 \
438         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA)                   \
439         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_HIGHMEM)               \
440         | 1 << (___GFP_MOVABLE | ___GFP_DMA32 | ___GFP_DMA | ___GFP_HIGHMEM)  \
441 )
442 
443 static inline enum zone_type gfp_zone(gfp_t flags)
444 {
445         enum zone_type z;
446         int bit = (__force int) (flags & GFP_ZONEMASK);
447 
448         z = (GFP_ZONE_TABLE >> (bit * GFP_ZONES_SHIFT)) &
449                                          ((1 << GFP_ZONES_SHIFT) - 1);
450         VM_BUG_ON((GFP_ZONE_BAD >> bit) & 1);
451         return z;
452 }
453 
454 /*
455  * There is only one page-allocator function, and two main namespaces to
456  * it. The alloc_page*() variants return 'struct page *' and as such
457  * can allocate highmem pages, the *get*page*() variants return
458  * virtual kernel addresses to the allocated page(s).
459  */
460 
461 static inline int gfp_zonelist(gfp_t flags)
462 {
463 #ifdef CONFIG_NUMA
464         if (unlikely(flags & __GFP_THISNODE))
465                 return ZONELIST_NOFALLBACK;
466 #endif
467         return ZONELIST_FALLBACK;
468 }
469 
470 /*
471  * We get the zone list from the current node and the gfp_mask.
472  * This zone list contains a maximum of MAXNODES*MAX_NR_ZONES zones.
473  * There are two zonelists per node, one for all zones with memory and
474  * one containing just zones from the node the zonelist belongs to.
475  *
476  * For the normal case of non-DISCONTIGMEM systems the NODE_DATA() gets
477  * optimized to &contig_page_data at compile-time.
478  */
479 static inline struct zonelist *node_zonelist(int nid, gfp_t flags)
480 {
481         return NODE_DATA(nid)->node_zonelists + gfp_zonelist(flags);
482 }
483 
484 #ifndef HAVE_ARCH_FREE_PAGE
485 static inline void arch_free_page(struct page *page, int order) { }
486 #endif
487 #ifndef HAVE_ARCH_ALLOC_PAGE
488 static inline void arch_alloc_page(struct page *page, int order) { }
489 #endif
490 #ifndef HAVE_ARCH_MAKE_PAGE_ACCESSIBLE
491 static inline int arch_make_page_accessible(struct page *page)
492 {
493         return 0;
494 }
495 #endif
496 
497 struct page *
498 __alloc_pages_nodemask(gfp_t gfp_mask, unsigned int order, int preferred_nid,
499                                                         nodemask_t *nodemask);
500 
501 static inline struct page *
502 __alloc_pages(gfp_t gfp_mask, unsigned int order, int preferred_nid)
503 {
504         return __alloc_pages_nodemask(gfp_mask, order, preferred_nid, NULL);
505 }
506 
507 /*
508  * Allocate pages, preferring the node given as nid. The node must be valid and
509  * online. For more general interface, see alloc_pages_node().
510  */
511 static inline struct page *
512 __alloc_pages_node(int nid, gfp_t gfp_mask, unsigned int order)
513 {
514         VM_BUG_ON(nid < 0 || nid >= MAX_NUMNODES);
515         VM_WARN_ON((gfp_mask & __GFP_THISNODE) && !node_online(nid));
516 
517         return __alloc_pages(gfp_mask, order, nid);
518 }
519 
520 /*
521  * Allocate pages, preferring the node given as nid. When nid == NUMA_NO_NODE,
522  * prefer the current CPU's closest node. Otherwise node must be valid and
523  * online.
524  */
525 static inline struct page *alloc_pages_node(int nid, gfp_t gfp_mask,
526                                                 unsigned int order)
527 {
528         if (nid == NUMA_NO_NODE)
529                 nid = numa_mem_id();
530 
531         return __alloc_pages_node(nid, gfp_mask, order);
532 }
533 
534 #ifdef CONFIG_NUMA
535 extern struct page *alloc_pages_current(gfp_t gfp_mask, unsigned order);
536 
537 static inline struct page *
538 alloc_pages(gfp_t gfp_mask, unsigned int order)
539 {
540         return alloc_pages_current(gfp_mask, order);
541 }
542 extern struct page *alloc_pages_vma(gfp_t gfp_mask, int order,
543                         struct vm_area_struct *vma, unsigned long addr,
544                         int node, bool hugepage);
545 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
546         alloc_pages_vma(gfp_mask, order, vma, addr, numa_node_id(), true)
547 #else
548 #define alloc_pages(gfp_mask, order) \
549                 alloc_pages_node(numa_node_id(), gfp_mask, order)
550 #define alloc_pages_vma(gfp_mask, order, vma, addr, node, false)\
551         alloc_pages(gfp_mask, order)
552 #define alloc_hugepage_vma(gfp_mask, vma, addr, order) \
553         alloc_pages(gfp_mask, order)
554 #endif
555 #define alloc_page(gfp_mask) alloc_pages(gfp_mask, 0)
556 #define alloc_page_vma(gfp_mask, vma, addr)                     \
557         alloc_pages_vma(gfp_mask, 0, vma, addr, numa_node_id(), false)
558 #define alloc_page_vma_node(gfp_mask, vma, addr, node)          \
559         alloc_pages_vma(gfp_mask, 0, vma, addr, node, false)
560 
561 extern unsigned long __get_free_pages(gfp_t gfp_mask, unsigned int order);
562 extern unsigned long get_zeroed_page(gfp_t gfp_mask);
563 
564 void *alloc_pages_exact(size_t size, gfp_t gfp_mask);
565 void free_pages_exact(void *virt, size_t size);
566 void * __meminit alloc_pages_exact_nid(int nid, size_t size, gfp_t gfp_mask);
567 
568 #define __get_free_page(gfp_mask) \
569                 __get_free_pages((gfp_mask), 0)
570 
571 #define __get_dma_pages(gfp_mask, order) \
572                 __get_free_pages((gfp_mask) | GFP_DMA, (order))
573 
574 extern void __free_pages(struct page *page, unsigned int order);
575 extern void free_pages(unsigned long addr, unsigned int order);
576 extern void free_unref_page(struct page *page);
577 extern void free_unref_page_list(struct list_head *list);
578 
579 struct page_frag_cache;
580 extern void __page_frag_cache_drain(struct page *page, unsigned int count);
581 extern void *page_frag_alloc(struct page_frag_cache *nc,
582                              unsigned int fragsz, gfp_t gfp_mask);
583 extern void page_frag_free(void *addr);
584 
585 #define __free_page(page) __free_pages((page), 0)
586 #define free_page(addr) free_pages((addr), 0)
587 
588 void page_alloc_init(void);
589 void drain_zone_pages(struct zone *zone, struct per_cpu_pages *pcp);
590 void drain_all_pages(struct zone *zone);
591 void drain_local_pages(struct zone *zone);
592 
593 void page_alloc_init_late(void);
594 
595 /*
596  * gfp_allowed_mask is set to GFP_BOOT_MASK during early boot to restrict what
597  * GFP flags are used before interrupts are enabled. Once interrupts are
598  * enabled, it is set to __GFP_BITS_MASK while the system is running. During
599  * hibernation, it is used by PM to avoid I/O during memory allocation while
600  * devices are suspended.
601  */
602 extern gfp_t gfp_allowed_mask;
603 
604 /* Returns true if the gfp_mask allows use of ALLOC_NO_WATERMARK */
605 bool gfp_pfmemalloc_allowed(gfp_t gfp_mask);
606 
607 extern void pm_restrict_gfp_mask(void);
608 extern void pm_restore_gfp_mask(void);
609 
610 #ifdef CONFIG_PM_SLEEP
611 extern bool pm_suspended_storage(void);
612 #else
613 static inline bool pm_suspended_storage(void)
614 {
615         return false;
616 }
617 #endif /* CONFIG_PM_SLEEP */
618 
619 #ifdef CONFIG_CONTIG_ALLOC
620 /* The below functions must be run on a range from a single zone. */
621 extern int alloc_contig_range(unsigned long start, unsigned long end,
622                               unsigned migratetype, gfp_t gfp_mask);
623 extern struct page *alloc_contig_pages(unsigned long nr_pages, gfp_t gfp_mask,
624                                        int nid, nodemask_t *nodemask);
625 #endif
626 void free_contig_range(unsigned long pfn, unsigned int nr_pages);
627 
628 #ifdef CONFIG_CMA
629 /* CMA stuff */
630 extern void init_cma_reserved_pageblock(struct page *page);
631 #endif
632 
633 #endif /* __LINUX_GFP_H */
634 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp