thp: transparent hugepage core
[linux-2.6-block.git] / include / linux / page-flags.h
CommitLineData
1da177e4
LT
1/*
2 * Macros for manipulating and testing page->flags
3 */
4
5#ifndef PAGE_FLAGS_H
6#define PAGE_FLAGS_H
7
f886ed44 8#include <linux/types.h>
9223b419 9#ifndef __GENERATING_BOUNDS_H
6d777953 10#include <linux/mm_types.h>
01fc0ac1 11#include <generated/bounds.h>
9223b419 12#endif /* !__GENERATING_BOUNDS_H */
f886ed44 13
1da177e4
LT
14/*
15 * Various page->flags bits:
16 *
17 * PG_reserved is set for special pages, which can never be swapped out. Some
18 * of them might not even exist (eg empty_bad_page)...
19 *
da6052f7
NP
20 * The PG_private bitflag is set on pagecache pages if they contain filesystem
21 * specific data (which is normally at page->private). It can be used by
22 * private allocations for its own usage.
1da177e4 23 *
da6052f7
NP
24 * During initiation of disk I/O, PG_locked is set. This bit is set before I/O
25 * and cleared when writeback _starts_ or when read _completes_. PG_writeback
26 * is set before writeback starts and cleared when it finishes.
27 *
28 * PG_locked also pins a page in pagecache, and blocks truncation of the file
29 * while it is held.
30 *
31 * page_waitqueue(page) is a wait queue of all tasks waiting for the page
32 * to become unlocked.
1da177e4
LT
33 *
34 * PG_uptodate tells whether the page's contents is valid. When a read
35 * completes, the page becomes uptodate, unless a disk I/O error happened.
36 *
da6052f7
NP
37 * PG_referenced, PG_reclaim are used for page reclaim for anonymous and
38 * file-backed pagecache (see mm/vmscan.c).
1da177e4
LT
39 *
40 * PG_error is set to indicate that an I/O error occurred on this page.
41 *
42 * PG_arch_1 is an architecture specific page state bit. The generic code
43 * guarantees that this bit is cleared for a page when it first is entered into
44 * the page cache.
45 *
46 * PG_highmem pages are not permanently mapped into the kernel virtual address
47 * space, they need to be kmapped separately for doing IO on the pages. The
48 * struct page (these bits with information) are always mapped into kernel
49 * address space...
da6052f7
NP
50 *
51 * PG_buddy is set to indicate that the page is free and in the buddy system
52 * (see mm/page_alloc.c).
53 *
d466f2fc
AK
54 * PG_hwpoison indicates that a page got corrupted in hardware and contains
55 * data with incorrect ECC bits that triggered a machine check. Accessing is
56 * not safe since it may cause another machine check. Don't touch!
1da177e4
LT
57 */
58
59/*
60 * Don't use the *_dontuse flags. Use the macros. Otherwise you'll break
91fc8ab3
AW
61 * locked- and dirty-page accounting.
62 *
63 * The page flags field is split into two parts, the main flags area
64 * which extends from the low bits upwards, and the fields area which
65 * extends from the high bits downwards.
66 *
67 * | FIELD | ... | FLAGS |
9223b419
CL
68 * N-1 ^ 0
69 * (NR_PAGEFLAGS)
91fc8ab3 70 *
9223b419
CL
71 * The fields area is reserved for fields mapping zone, node (for NUMA) and
72 * SPARSEMEM section (for variants of SPARSEMEM that require section ids like
73 * SPARSEMEM_EXTREME with !SPARSEMEM_VMEMMAP).
1da177e4 74 */
e2683181
CL
75enum pageflags {
76 PG_locked, /* Page is locked. Don't touch. */
77 PG_error,
78 PG_referenced,
79 PG_uptodate,
80 PG_dirty,
81 PG_lru,
82 PG_active,
83 PG_slab,
84 PG_owner_priv_1, /* Owner use. If pagecache, fs may use*/
e2683181
CL
85 PG_arch_1,
86 PG_reserved,
87 PG_private, /* If pagecache, has fs-private data */
266cf658 88 PG_private_2, /* If pagecache, has fs aux data */
e2683181 89 PG_writeback, /* Page is under writeback */
e20b8cca
CL
90#ifdef CONFIG_PAGEFLAGS_EXTENDED
91 PG_head, /* A head page */
92 PG_tail, /* A tail page */
93#else
e2683181 94 PG_compound, /* A compound page */
e20b8cca 95#endif
e2683181
CL
96 PG_swapcache, /* Swap page: swp_entry_t in private */
97 PG_mappedtodisk, /* Has blocks allocated on-disk */
98 PG_reclaim, /* To be reclaimed asap */
e2683181 99 PG_buddy, /* Page is free, on buddy lists */
b2e18538 100 PG_swapbacked, /* Page is backed by RAM/swap */
894bc310 101 PG_unevictable, /* Page is "unevictable" */
af8e3354 102#ifdef CONFIG_MMU
b291f000 103 PG_mlocked, /* Page is vma mlocked */
894bc310 104#endif
46cf98cd 105#ifdef CONFIG_ARCH_USES_PG_UNCACHED
602c4d11 106 PG_uncached, /* Page has been mapped as uncached */
d466f2fc
AK
107#endif
108#ifdef CONFIG_MEMORY_FAILURE
109 PG_hwpoison, /* hardware poisoned page. Don't touch */
e9da73d6
AA
110#endif
111#ifdef CONFIG_TRANSPARENT_HUGEPAGE
112 PG_compound_lock,
f886ed44 113#endif
0cad47cf
AW
114 __NR_PAGEFLAGS,
115
116 /* Filesystems */
117 PG_checked = PG_owner_priv_1,
118
266cf658
DH
119 /* Two page bits are conscripted by FS-Cache to maintain local caching
120 * state. These bits are set on pages belonging to the netfs's inodes
121 * when those inodes are being locally cached.
122 */
123 PG_fscache = PG_private_2, /* page backed by cache */
124
0cad47cf
AW
125 /* XEN */
126 PG_pinned = PG_owner_priv_1,
127 PG_savepinned = PG_dirty,
8a38082d 128
9023cb7e 129 /* SLOB */
9023cb7e
AW
130 PG_slob_free = PG_private,
131
8a38082d
AW
132 /* SLUB */
133 PG_slub_frozen = PG_active,
e2683181 134};
1da177e4 135
9223b419
CL
136#ifndef __GENERATING_BOUNDS_H
137
f94a62e9
CL
138/*
139 * Macros to create function definitions for page flags
140 */
141#define TESTPAGEFLAG(uname, lname) \
142static inline int Page##uname(struct page *page) \
143 { return test_bit(PG_##lname, &page->flags); }
144
145#define SETPAGEFLAG(uname, lname) \
146static inline void SetPage##uname(struct page *page) \
147 { set_bit(PG_##lname, &page->flags); }
148
149#define CLEARPAGEFLAG(uname, lname) \
150static inline void ClearPage##uname(struct page *page) \
151 { clear_bit(PG_##lname, &page->flags); }
152
153#define __SETPAGEFLAG(uname, lname) \
154static inline void __SetPage##uname(struct page *page) \
155 { __set_bit(PG_##lname, &page->flags); }
156
157#define __CLEARPAGEFLAG(uname, lname) \
158static inline void __ClearPage##uname(struct page *page) \
159 { __clear_bit(PG_##lname, &page->flags); }
160
161#define TESTSETFLAG(uname, lname) \
162static inline int TestSetPage##uname(struct page *page) \
163 { return test_and_set_bit(PG_##lname, &page->flags); }
164
165#define TESTCLEARFLAG(uname, lname) \
166static inline int TestClearPage##uname(struct page *page) \
167 { return test_and_clear_bit(PG_##lname, &page->flags); }
168
451ea25d
JW
169#define __TESTCLEARFLAG(uname, lname) \
170static inline int __TestClearPage##uname(struct page *page) \
171 { return __test_and_clear_bit(PG_##lname, &page->flags); }
f94a62e9
CL
172
173#define PAGEFLAG(uname, lname) TESTPAGEFLAG(uname, lname) \
174 SETPAGEFLAG(uname, lname) CLEARPAGEFLAG(uname, lname)
175
176#define __PAGEFLAG(uname, lname) TESTPAGEFLAG(uname, lname) \
177 __SETPAGEFLAG(uname, lname) __CLEARPAGEFLAG(uname, lname)
178
ec7cade8
CL
179#define PAGEFLAG_FALSE(uname) \
180static inline int Page##uname(struct page *page) \
181 { return 0; }
182
f94a62e9
CL
183#define TESTSCFLAG(uname, lname) \
184 TESTSETFLAG(uname, lname) TESTCLEARFLAG(uname, lname)
185
8a7a8544
LS
186#define SETPAGEFLAG_NOOP(uname) \
187static inline void SetPage##uname(struct page *page) { }
188
189#define CLEARPAGEFLAG_NOOP(uname) \
190static inline void ClearPage##uname(struct page *page) { }
191
192#define __CLEARPAGEFLAG_NOOP(uname) \
193static inline void __ClearPage##uname(struct page *page) { }
194
195#define TESTCLEARFLAG_FALSE(uname) \
196static inline int TestClearPage##uname(struct page *page) { return 0; }
197
451ea25d
JW
198#define __TESTCLEARFLAG_FALSE(uname) \
199static inline int __TestClearPage##uname(struct page *page) { return 0; }
200
6a1e7f77
CL
201struct page; /* forward declaration */
202
03fb3d2a 203TESTPAGEFLAG(Locked, locked) TESTSETFLAG(Locked, locked)
212260aa 204PAGEFLAG(Error, error) TESTCLEARFLAG(Error, error)
6a1e7f77
CL
205PAGEFLAG(Referenced, referenced) TESTCLEARFLAG(Referenced, referenced)
206PAGEFLAG(Dirty, dirty) TESTSCFLAG(Dirty, dirty) __CLEARPAGEFLAG(Dirty, dirty)
207PAGEFLAG(LRU, lru) __CLEARPAGEFLAG(LRU, lru)
208PAGEFLAG(Active, active) __CLEARPAGEFLAG(Active, active)
894bc310 209 TESTCLEARFLAG(Active, active)
6a1e7f77 210__PAGEFLAG(Slab, slab)
0cad47cf
AW
211PAGEFLAG(Checked, checked) /* Used by some filesystems */
212PAGEFLAG(Pinned, pinned) TESTSCFLAG(Pinned, pinned) /* Xen */
213PAGEFLAG(SavePinned, savepinned); /* Xen */
6a1e7f77 214PAGEFLAG(Reserved, reserved) __CLEARPAGEFLAG(Reserved, reserved)
b2e18538 215PAGEFLAG(SwapBacked, swapbacked) __CLEARPAGEFLAG(SwapBacked, swapbacked)
6a1e7f77 216
9023cb7e
AW
217__PAGEFLAG(SlobFree, slob_free)
218
8a38082d 219__PAGEFLAG(SlubFrozen, slub_frozen)
8a38082d 220
266cf658
DH
221/*
222 * Private page markings that may be used by the filesystem that owns the page
223 * for its own purposes.
224 * - PG_private and PG_private_2 cause releasepage() and co to be invoked
225 */
226PAGEFLAG(Private, private) __SETPAGEFLAG(Private, private)
227 __CLEARPAGEFLAG(Private, private)
228PAGEFLAG(Private2, private_2) TESTSCFLAG(Private2, private_2)
229PAGEFLAG(OwnerPriv1, owner_priv_1) TESTCLEARFLAG(OwnerPriv1, owner_priv_1)
230
6a1e7f77
CL
231/*
232 * Only test-and-set exist for PG_writeback. The unconditional operators are
233 * risky: they bypass page accounting.
234 */
235TESTPAGEFLAG(Writeback, writeback) TESTSCFLAG(Writeback, writeback)
236__PAGEFLAG(Buddy, buddy)
237PAGEFLAG(MappedToDisk, mappedtodisk)
238
239/* PG_readahead is only used for file reads; PG_reclaim is only for writes */
240PAGEFLAG(Reclaim, reclaim) TESTCLEARFLAG(Reclaim, reclaim)
0a128b2b 241PAGEFLAG(Readahead, reclaim) /* Reminder to do async read-ahead */
6a1e7f77
CL
242
243#ifdef CONFIG_HIGHMEM
1da177e4 244/*
6a1e7f77
CL
245 * Must use a macro here due to header dependency issues. page_zone() is not
246 * available at this point.
1da177e4 247 */
0a128b2b 248#define PageHighMem(__p) is_highmem(page_zone(__p))
6a1e7f77 249#else
ec7cade8 250PAGEFLAG_FALSE(HighMem)
6a1e7f77
CL
251#endif
252
253#ifdef CONFIG_SWAP
254PAGEFLAG(SwapCache, swapcache)
255#else
ec7cade8 256PAGEFLAG_FALSE(SwapCache)
6d91add0 257 SETPAGEFLAG_NOOP(SwapCache) CLEARPAGEFLAG_NOOP(SwapCache)
6a1e7f77
CL
258#endif
259
894bc310
LS
260PAGEFLAG(Unevictable, unevictable) __CLEARPAGEFLAG(Unevictable, unevictable)
261 TESTCLEARFLAG(Unevictable, unevictable)
b291f000 262
af8e3354 263#ifdef CONFIG_MMU
b291f000 264PAGEFLAG(Mlocked, mlocked) __CLEARPAGEFLAG(Mlocked, mlocked)
451ea25d 265 TESTSCFLAG(Mlocked, mlocked) __TESTCLEARFLAG(Mlocked, mlocked)
894bc310 266#else
451ea25d
JW
267PAGEFLAG_FALSE(Mlocked) SETPAGEFLAG_NOOP(Mlocked)
268 TESTCLEARFLAG_FALSE(Mlocked) __TESTCLEARFLAG_FALSE(Mlocked)
894bc310
LS
269#endif
270
46cf98cd 271#ifdef CONFIG_ARCH_USES_PG_UNCACHED
6a1e7f77 272PAGEFLAG(Uncached, uncached)
602c4d11 273#else
ec7cade8 274PAGEFLAG_FALSE(Uncached)
6a1e7f77 275#endif
1da177e4 276
d466f2fc
AK
277#ifdef CONFIG_MEMORY_FAILURE
278PAGEFLAG(HWPoison, hwpoison)
847ce401 279TESTSCFLAG(HWPoison, hwpoison)
d466f2fc
AK
280#define __PG_HWPOISON (1UL << PG_hwpoison)
281#else
282PAGEFLAG_FALSE(HWPoison)
283#define __PG_HWPOISON 0
284#endif
285
1a9b5b7f
WF
286u64 stable_page_flags(struct page *page);
287
0ed361de
NP
288static inline int PageUptodate(struct page *page)
289{
290 int ret = test_bit(PG_uptodate, &(page)->flags);
291
292 /*
293 * Must ensure that the data we read out of the page is loaded
294 * _after_ we've loaded page->flags to check for PageUptodate.
295 * We can skip the barrier if the page is not uptodate, because
296 * we wouldn't be reading anything from it.
297 *
298 * See SetPageUptodate() for the other side of the story.
299 */
300 if (ret)
301 smp_rmb();
302
303 return ret;
304}
305
306static inline void __SetPageUptodate(struct page *page)
307{
308 smp_wmb();
309 __set_bit(PG_uptodate, &(page)->flags);
0ed361de
NP
310}
311
2dcea57a
HC
312static inline void SetPageUptodate(struct page *page)
313{
0ed361de 314#ifdef CONFIG_S390
2dcea57a 315 if (!test_and_set_bit(PG_uptodate, &page->flags))
e2b8d7af 316 page_clear_dirty(page, 0);
f6ac2354 317#else
0ed361de
NP
318 /*
319 * Memory barrier must be issued before setting the PG_uptodate bit,
320 * so that all previous stores issued in order to bring the page
321 * uptodate are actually visible before PageUptodate becomes true.
322 *
323 * s390 doesn't need an explicit smp_wmb here because the test and
324 * set bit already provides full barriers.
325 */
326 smp_wmb();
327 set_bit(PG_uptodate, &(page)->flags);
1da177e4 328#endif
0ed361de
NP
329}
330
6a1e7f77 331CLEARPAGEFLAG(Uptodate, uptodate)
1da177e4 332
6a1e7f77 333extern void cancel_dirty_page(struct page *page, unsigned int account_size);
d77c2d7c 334
6a1e7f77
CL
335int test_clear_page_writeback(struct page *page);
336int test_set_page_writeback(struct page *page);
1da177e4 337
6a1e7f77
CL
338static inline void set_page_writeback(struct page *page)
339{
340 test_set_page_writeback(page);
341}
1da177e4 342
e20b8cca
CL
343#ifdef CONFIG_PAGEFLAGS_EXTENDED
344/*
345 * System with lots of page flags available. This allows separate
346 * flags for PageHead() and PageTail() checks of compound pages so that bit
347 * tests can be used in performance sensitive paths. PageCompound is
348 * generally not used in hot code paths.
349 */
4e6af67e 350__PAGEFLAG(Head, head) CLEARPAGEFLAG(Head, head)
e20b8cca
CL
351__PAGEFLAG(Tail, tail)
352
353static inline int PageCompound(struct page *page)
354{
355 return page->flags & ((1L << PG_head) | (1L << PG_tail));
356
357}
4e6af67e
AA
358#ifdef CONFIG_TRANSPARENT_HUGEPAGE
359static inline void ClearPageCompound(struct page *page)
360{
361 BUG_ON(!PageHead(page));
362 ClearPageHead(page);
363}
364#endif
e20b8cca
CL
365#else
366/*
367 * Reduce page flag use as much as possible by overlapping
368 * compound page flags with the flags used for page cache pages. Possible
369 * because PageCompound is always set for compound pages and not for
370 * pages on the LRU and/or pagecache.
371 */
6a1e7f77
CL
372TESTPAGEFLAG(Compound, compound)
373__PAGEFLAG(Head, compound)
1da177e4 374
d85f3385 375/*
6d777953 376 * PG_reclaim is used in combination with PG_compound to mark the
6a1e7f77
CL
377 * head and tail of a compound page. This saves one page flag
378 * but makes it impossible to use compound pages for the page cache.
379 * The PG_reclaim bit would have to be used for reclaim or readahead
380 * if compound pages enter the page cache.
6d777953
CL
381 *
382 * PG_compound & PG_reclaim => Tail page
383 * PG_compound & ~PG_reclaim => Head page
d85f3385 384 */
6d777953
CL
385#define PG_head_tail_mask ((1L << PG_compound) | (1L << PG_reclaim))
386
6a1e7f77
CL
387static inline int PageTail(struct page *page)
388{
389 return ((page->flags & PG_head_tail_mask) == PG_head_tail_mask);
390}
6d777953
CL
391
392static inline void __SetPageTail(struct page *page)
393{
394 page->flags |= PG_head_tail_mask;
395}
396
397static inline void __ClearPageTail(struct page *page)
398{
399 page->flags &= ~PG_head_tail_mask;
400}
401
4e6af67e
AA
402#ifdef CONFIG_TRANSPARENT_HUGEPAGE
403static inline void ClearPageCompound(struct page *page)
404{
405 BUG_ON((page->flags & PG_head_tail_mask) != (1 << PG_compound));
406 clear_bit(PG_compound, &page->flags);
407}
408#endif
409
e20b8cca 410#endif /* !PAGEFLAGS_EXTENDED */
dfa7e20c 411
936a5fe6 412#ifdef CONFIG_TRANSPARENT_HUGEPAGE
71e3aac0
AA
413/*
414 * PageHuge() only returns true for hugetlbfs pages, but not for
415 * normal or transparent huge pages.
416 *
417 * PageTransHuge() returns true for both transparent huge and
418 * hugetlbfs pages, but not normal pages. PageTransHuge() can only be
419 * called only in the core VM paths where hugetlbfs pages can't exist.
420 */
421static inline int PageTransHuge(struct page *page)
422{
423 VM_BUG_ON(PageTail(page));
424 return PageHead(page);
425}
426
936a5fe6
AA
427static inline int PageTransCompound(struct page *page)
428{
429 return PageCompound(page);
430}
71e3aac0 431
936a5fe6 432#else
71e3aac0
AA
433
434static inline int PageTransHuge(struct page *page)
435{
436 return 0;
437}
438
936a5fe6
AA
439static inline int PageTransCompound(struct page *page)
440{
441 return 0;
442}
443#endif
444
af8e3354 445#ifdef CONFIG_MMU
33925b25
DH
446#define __PG_MLOCKED (1 << PG_mlocked)
447#else
b291f000 448#define __PG_MLOCKED 0
894bc310
LS
449#endif
450
e9da73d6
AA
451#ifdef CONFIG_TRANSPARENT_HUGEPAGE
452#define __PG_COMPOUND_LOCK (1 << PG_compound_lock)
453#else
454#define __PG_COMPOUND_LOCK 0
455#endif
456
dfa7e20c
RA
457/*
458 * Flags checked when a page is freed. Pages being freed should not have
459 * these flags set. It they are, there is a problem.
460 */
79f4b7bf 461#define PAGE_FLAGS_CHECK_AT_FREE \
266cf658
DH
462 (1 << PG_lru | 1 << PG_locked | \
463 1 << PG_private | 1 << PG_private_2 | \
464 1 << PG_buddy | 1 << PG_writeback | 1 << PG_reserved | \
465 1 << PG_slab | 1 << PG_swapcache | 1 << PG_active | \
e9da73d6
AA
466 1 << PG_unevictable | __PG_MLOCKED | __PG_HWPOISON | \
467 __PG_COMPOUND_LOCK)
dfa7e20c
RA
468
469/*
470 * Flags checked when a page is prepped for return by the page allocator.
79f4b7bf
HD
471 * Pages being prepped should not have any flags set. It they are set,
472 * there has been a kernel bug or struct page corruption.
dfa7e20c 473 */
79f4b7bf 474#define PAGE_FLAGS_CHECK_AT_PREP ((1 << NR_PAGEFLAGS) - 1)
dfa7e20c 475
edcf4748
JW
476#define PAGE_FLAGS_PRIVATE \
477 (1 << PG_private | 1 << PG_private_2)
266cf658
DH
478/**
479 * page_has_private - Determine if page has private stuff
480 * @page: The page to be checked
481 *
482 * Determine if a page has private stuff, indicating that release routines
483 * should be invoked upon it.
484 */
edcf4748
JW
485static inline int page_has_private(struct page *page)
486{
487 return !!(page->flags & PAGE_FLAGS_PRIVATE);
488}
489
490#endif /* !__GENERATING_BOUNDS_H */
266cf658 491
1da177e4 492#endif /* PAGE_FLAGS_H */