Skip to content

Navigation Menu

Sign in

Search code, repositories, users, issues, pull requests...

Provide feedback

We read every piece of feedback, and take your input very seriously.

Saved searches

Use saved searches to filter your results more quickly
Appearance settings

Commit faeedbc

Browse filesBrowse files
macdicemacdice
committed
Introduce PG_IO_ALIGN_SIZE and align all I/O buffers.
In order to have the option to use O_DIRECT/FILE_FLAG_NO_BUFFERING in a later commit, we need the addresses of user space buffers to be well aligned. The exact requirements vary by OS and file system (typically sectors and/or memory pages). The address alignment size is set to 4096, which is enough for currently known systems: it matches modern sectors and common memory page size. There is no standard governing O_DIRECT's requirements so we might eventually have to reconsider this with more information from the field or future systems. Aligning I/O buffers on memory pages is also known to improve regular buffered I/O performance. Three classes of I/O buffers for regular data pages are adjusted: (1) Heap buffers are now allocated with the new palloc_aligned() or MemoryContextAllocAligned() functions introduced by commit 439f617. (2) Stack buffers now use a new struct PGIOAlignedBlock to respect PG_IO_ALIGN_SIZE, if possible with this compiler. (3) The buffer pool is also aligned in shared memory. WAL buffers were already aligned on XLOG_BLCKSZ. It's possible for XLOG_BLCKSZ to be configured smaller than PG_IO_ALIGNED_SIZE and thus for O_DIRECT WAL writes to fail to be well aligned, but that's a pre-existing condition and will be addressed by a later commit. BufFiles are not yet addressed (there's no current plan to use O_DIRECT for those, but they could potentially get some incidental speedup even in plain buffered I/O operations through better alignment). If we can't align stack objects suitably using the compiler extensions we know about, we disable the use of O_DIRECT by setting PG_O_DIRECT to 0. This avoids the need to consider systems that have O_DIRECT but can't align stack objects the way we want; such systems could in theory be supported with more work but we don't currently know of any such machines, so it's easier to pretend there is no O_DIRECT support instead. That's an existing and tested class of system. Add assertions that all buffers passed into smgrread(), smgrwrite() and smgrextend() are correctly aligned, unless PG_O_DIRECT is 0 (= stack alignment tricks may be unavailable) or the block size has been set too small to allow arrays of buffers to be all aligned. Author: Thomas Munro <thomas.munro@gmail.com> Author: Andres Freund <andres@anarazel.de> Reviewed-by: Justin Pryzby <pryzby@telsasoft.com> Discussion: https://postgr.es/m/CA+hUKGK1X532hYqJ_MzFWt0n1zt8trz980D79WbjwnT-yYLZpg@mail.gmail.com
1 parent d73c285 commit faeedbc
Copy full SHA for faeedbc

File tree

26 files changed

+108
-45
lines changed
Filter options

26 files changed

+108
-45
lines changed

‎contrib/bloom/blinsert.c

Copy file name to clipboardExpand all lines: contrib/bloom/blinsert.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -166,7 +166,7 @@ blbuildempty(Relation index)
166166
Page metapage;
167167

168168
/* Construct metapage. */
169-
metapage = (Page) palloc(BLCKSZ);
169+
metapage = (Page) palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
170170
BloomFillMetapage(index, metapage);
171171

172172
/*

‎contrib/pg_prewarm/pg_prewarm.c

Copy file name to clipboardExpand all lines: contrib/pg_prewarm/pg_prewarm.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -36,7 +36,7 @@ typedef enum
3636
PREWARM_BUFFER
3737
} PrewarmType;
3838

39-
static PGAlignedBlock blockbuffer;
39+
static PGIOAlignedBlock blockbuffer;
4040

4141
/*
4242
* pg_prewarm(regclass, mode text, fork text,

‎src/backend/access/gist/gistbuild.c

Copy file name to clipboardExpand all lines: src/backend/access/gist/gistbuild.c
+5-4Lines changed: 5 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -415,7 +415,7 @@ gist_indexsortbuild(GISTBuildState *state)
415415
* Write an empty page as a placeholder for the root page. It will be
416416
* replaced with the real root page at the end.
417417
*/
418-
page = palloc0(BLCKSZ);
418+
page = palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, MCXT_ALLOC_ZERO);
419419
smgrextend(RelationGetSmgr(state->indexrel), MAIN_FORKNUM, GIST_ROOT_BLKNO,
420420
page, true);
421421
state->pages_allocated++;
@@ -509,7 +509,8 @@ gist_indexsortbuild_levelstate_add(GISTBuildState *state,
509509
levelstate->current_page++;
510510

511511
if (levelstate->pages[levelstate->current_page] == NULL)
512-
levelstate->pages[levelstate->current_page] = palloc(BLCKSZ);
512+
levelstate->pages[levelstate->current_page] =
513+
palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
513514

514515
newPage = levelstate->pages[levelstate->current_page];
515516
gistinitpage(newPage, old_page_flags);
@@ -579,7 +580,7 @@ gist_indexsortbuild_levelstate_flush(GISTBuildState *state,
579580

580581
/* Create page and copy data */
581582
data = (char *) (dist->list);
582-
target = palloc0(BLCKSZ);
583+
target = palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, MCXT_ALLOC_ZERO);
583584
gistinitpage(target, isleaf ? F_LEAF : 0);
584585
for (int i = 0; i < dist->block.num; i++)
585586
{
@@ -630,7 +631,7 @@ gist_indexsortbuild_levelstate_flush(GISTBuildState *state,
630631
if (parent == NULL)
631632
{
632633
parent = palloc0(sizeof(GistSortedBuildLevelState));
633-
parent->pages[0] = (Page) palloc(BLCKSZ);
634+
parent->pages[0] = (Page) palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
634635
parent->parent = NULL;
635636
gistinitpage(parent->pages[0], 0);
636637

‎src/backend/access/hash/hashpage.c

Copy file name to clipboardExpand all lines: src/backend/access/hash/hashpage.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -992,7 +992,7 @@ static bool
992992
_hash_alloc_buckets(Relation rel, BlockNumber firstblock, uint32 nblocks)
993993
{
994994
BlockNumber lastblock;
995-
PGAlignedBlock zerobuf;
995+
PGIOAlignedBlock zerobuf;
996996
Page page;
997997
HashPageOpaque ovflopaque;
998998

‎src/backend/access/heap/rewriteheap.c

Copy file name to clipboardExpand all lines: src/backend/access/heap/rewriteheap.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -255,7 +255,7 @@ begin_heap_rewrite(Relation old_heap, Relation new_heap, TransactionId oldest_xm
255255

256256
state->rs_old_rel = old_heap;
257257
state->rs_new_rel = new_heap;
258-
state->rs_buffer = (Page) palloc(BLCKSZ);
258+
state->rs_buffer = (Page) palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
259259
/* new_heap needn't be empty, just locked */
260260
state->rs_blockno = RelationGetNumberOfBlocks(new_heap);
261261
state->rs_buffer_valid = false;

‎src/backend/access/nbtree/nbtree.c

Copy file name to clipboardExpand all lines: src/backend/access/nbtree/nbtree.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -154,7 +154,7 @@ btbuildempty(Relation index)
154154
Page metapage;
155155

156156
/* Construct metapage. */
157-
metapage = (Page) palloc(BLCKSZ);
157+
metapage = (Page) palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
158158
_bt_initmetapage(metapage, P_NONE, 0, _bt_allequalimage(index, false));
159159

160160
/*

‎src/backend/access/nbtree/nbtsort.c

Copy file name to clipboardExpand all lines: src/backend/access/nbtree/nbtsort.c
+5-3Lines changed: 5 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -619,7 +619,7 @@ _bt_blnewpage(uint32 level)
619619
Page page;
620620
BTPageOpaque opaque;
621621

622-
page = (Page) palloc(BLCKSZ);
622+
page = (Page) palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
623623

624624
/* Zero the page and set up standard page header info */
625625
_bt_pageinit(page, BLCKSZ);
@@ -660,7 +660,9 @@ _bt_blwritepage(BTWriteState *wstate, Page page, BlockNumber blkno)
660660
while (blkno > wstate->btws_pages_written)
661661
{
662662
if (!wstate->btws_zeropage)
663-
wstate->btws_zeropage = (Page) palloc0(BLCKSZ);
663+
wstate->btws_zeropage = (Page) palloc_aligned(BLCKSZ,
664+
PG_IO_ALIGN_SIZE,
665+
MCXT_ALLOC_ZERO);
664666
/* don't set checksum for all-zero page */
665667
smgrextend(RelationGetSmgr(wstate->index), MAIN_FORKNUM,
666668
wstate->btws_pages_written++,
@@ -1170,7 +1172,7 @@ _bt_uppershutdown(BTWriteState *wstate, BTPageState *state)
11701172
* set to point to "P_NONE"). This changes the index to the "valid" state
11711173
* by filling in a valid magic number in the metapage.
11721174
*/
1173-
metapage = (Page) palloc(BLCKSZ);
1175+
metapage = (Page) palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
11741176
_bt_initmetapage(metapage, rootblkno, rootlevel,
11751177
wstate->inskey->allequalimage);
11761178
_bt_blwritepage(wstate, metapage, BTREE_METAPAGE);

‎src/backend/access/spgist/spginsert.c

Copy file name to clipboardExpand all lines: src/backend/access/spgist/spginsert.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -158,7 +158,7 @@ spgbuildempty(Relation index)
158158
Page page;
159159

160160
/* Construct metapage. */
161-
page = (Page) palloc(BLCKSZ);
161+
page = (Page) palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE, 0);
162162
SpGistInitMetapage(page);
163163

164164
/*

‎src/backend/access/transam/generic_xlog.c

Copy file name to clipboardExpand all lines: src/backend/access/transam/generic_xlog.c
+9-4Lines changed: 9 additions & 4 deletions
Original file line numberDiff line numberDiff line change
@@ -58,14 +58,17 @@ typedef struct
5858
char delta[MAX_DELTA_SIZE]; /* delta between page images */
5959
} PageData;
6060

61-
/* State of generic xlog record construction */
61+
/*
62+
* State of generic xlog record construction. Must be allocated at an I/O
63+
* aligned address.
64+
*/
6265
struct GenericXLogState
6366
{
67+
/* Page images (properly aligned, must be first) */
68+
PGIOAlignedBlock images[MAX_GENERIC_XLOG_PAGES];
6469
/* Info about each page, see above */
6570
PageData pages[MAX_GENERIC_XLOG_PAGES];
6671
bool isLogged;
67-
/* Page images (properly aligned) */
68-
PGAlignedBlock images[MAX_GENERIC_XLOG_PAGES];
6972
};
7073

7174
static void writeFragment(PageData *pageData, OffsetNumber offset,
@@ -269,7 +272,9 @@ GenericXLogStart(Relation relation)
269272
GenericXLogState *state;
270273
int i;
271274

272-
state = (GenericXLogState *) palloc(sizeof(GenericXLogState));
275+
state = (GenericXLogState *) palloc_aligned(sizeof(GenericXLogState),
276+
PG_IO_ALIGN_SIZE,
277+
0);
273278
state->isLogged = RelationNeedsWAL(relation);
274279

275280
for (i = 0; i < MAX_GENERIC_XLOG_PAGES; i++)

‎src/backend/access/transam/xlog.c

Copy file name to clipboardExpand all lines: src/backend/access/transam/xlog.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4506,7 +4506,7 @@ XLOGShmemSize(void)
45064506
/* xlblocks array */
45074507
size = add_size(size, mul_size(sizeof(XLogRecPtr), XLOGbuffers));
45084508
/* extra alignment padding for XLOG I/O buffers */
4509-
size = add_size(size, XLOG_BLCKSZ);
4509+
size = add_size(size, Max(XLOG_BLCKSZ, PG_IO_ALIGN_SIZE));
45104510
/* and the buffers themselves */
45114511
size = add_size(size, mul_size(XLOG_BLCKSZ, XLOGbuffers));
45124512

‎src/backend/catalog/storage.c

Copy file name to clipboardExpand all lines: src/backend/catalog/storage.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -451,7 +451,7 @@ void
451451
RelationCopyStorage(SMgrRelation src, SMgrRelation dst,
452452
ForkNumber forkNum, char relpersistence)
453453
{
454-
PGAlignedBlock buf;
454+
PGIOAlignedBlock buf;
455455
Page page;
456456
bool use_wal;
457457
bool copying_initfork;

‎src/backend/storage/buffer/buf_init.c

Copy file name to clipboardExpand all lines: src/backend/storage/buffer/buf_init.c
+7-3Lines changed: 7 additions & 3 deletions
Original file line numberDiff line numberDiff line change
@@ -78,9 +78,12 @@ InitBufferPool(void)
7878
NBuffers * sizeof(BufferDescPadded),
7979
&foundDescs);
8080

81+
/* Align buffer pool on IO page size boundary. */
8182
BufferBlocks = (char *)
82-
ShmemInitStruct("Buffer Blocks",
83-
NBuffers * (Size) BLCKSZ, &foundBufs);
83+
TYPEALIGN(PG_IO_ALIGN_SIZE,
84+
ShmemInitStruct("Buffer Blocks",
85+
NBuffers * (Size) BLCKSZ + PG_IO_ALIGN_SIZE,
86+
&foundBufs));
8487

8588
/* Align condition variables to cacheline boundary. */
8689
BufferIOCVArray = (ConditionVariableMinimallyPadded *)
@@ -163,7 +166,8 @@ BufferShmemSize(void)
163166
/* to allow aligning buffer descriptors */
164167
size = add_size(size, PG_CACHE_LINE_SIZE);
165168

166-
/* size of data pages */
169+
/* size of data pages, plus alignment padding */
170+
size = add_size(size, PG_IO_ALIGN_SIZE);
167171
size = add_size(size, mul_size(NBuffers, BLCKSZ));
168172

169173
/* size of stuff controlled by freelist.c */

‎src/backend/storage/buffer/bufmgr.c

Copy file name to clipboardExpand all lines: src/backend/storage/buffer/bufmgr.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -4250,7 +4250,7 @@ RelationCopyStorageUsingBuffer(RelFileLocator srclocator,
42504250
bool use_wal;
42514251
BlockNumber nblocks;
42524252
BlockNumber blkno;
4253-
PGAlignedBlock buf;
4253+
PGIOAlignedBlock buf;
42544254
BufferAccessStrategy bstrategy_src;
42554255
BufferAccessStrategy bstrategy_dst;
42564256

‎src/backend/storage/buffer/localbuf.c

Copy file name to clipboardExpand all lines: src/backend/storage/buffer/localbuf.c
+5-2Lines changed: 5 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -744,8 +744,11 @@ GetLocalBufferStorage(void)
744744
/* And don't overflow MaxAllocSize, either */
745745
num_bufs = Min(num_bufs, MaxAllocSize / BLCKSZ);
746746

747-
cur_block = (char *) MemoryContextAlloc(LocalBufferContext,
748-
num_bufs * BLCKSZ);
747+
/* Buffers should be I/O aligned. */
748+
cur_block = (char *)
749+
TYPEALIGN(PG_IO_ALIGN_SIZE,
750+
MemoryContextAlloc(LocalBufferContext,
751+
num_bufs * BLCKSZ + PG_IO_ALIGN_SIZE));
749752
next_buf_in_block = 0;
750753
num_bufs_in_block = num_bufs;
751754
}

‎src/backend/storage/file/buffile.c

Copy file name to clipboardExpand all lines: src/backend/storage/file/buffile.c
+6Lines changed: 6 additions & 0 deletions
Original file line numberDiff line numberDiff line change
@@ -95,6 +95,12 @@ struct BufFile
9595
off_t curOffset; /* offset part of current pos */
9696
int pos; /* next read/write position in buffer */
9797
int nbytes; /* total # of valid bytes in buffer */
98+
99+
/*
100+
* XXX Should ideally us PGIOAlignedBlock, but might need a way to avoid
101+
* wasting per-file alignment padding when some users create many
102+
* files.
103+
*/
98104
PGAlignedBlock buffer;
99105
};
100106

‎src/backend/storage/page/bufpage.c

Copy file name to clipboardExpand all lines: src/backend/storage/page/bufpage.c
+4-1Lines changed: 4 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1522,7 +1522,10 @@ PageSetChecksumCopy(Page page, BlockNumber blkno)
15221522
* and second to avoid wasting space in processes that never call this.
15231523
*/
15241524
if (pageCopy == NULL)
1525-
pageCopy = MemoryContextAlloc(TopMemoryContext, BLCKSZ);
1525+
pageCopy = MemoryContextAllocAligned(TopMemoryContext,
1526+
BLCKSZ,
1527+
PG_IO_ALIGN_SIZE,
1528+
0);
15261529

15271530
memcpy(pageCopy, (char *) page, BLCKSZ);
15281531
((PageHeader) pageCopy)->pd_checksum = pg_checksum_page(pageCopy, blkno);

‎src/backend/storage/smgr/md.c

Copy file name to clipboardExpand all lines: src/backend/storage/smgr/md.c
+14-1Lines changed: 14 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -453,6 +453,10 @@ mdextend(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
453453
int nbytes;
454454
MdfdVec *v;
455455

456+
/* If this build supports direct I/O, the buffer must be I/O aligned. */
457+
if (PG_O_DIRECT != 0 && PG_IO_ALIGN_SIZE <= BLCKSZ)
458+
Assert((uintptr_t) buffer == TYPEALIGN(PG_IO_ALIGN_SIZE, buffer));
459+
456460
/* This assert is too expensive to have on normally ... */
457461
#ifdef CHECK_WRITE_VS_EXTEND
458462
Assert(blocknum >= mdnblocks(reln, forknum));
@@ -783,6 +787,10 @@ mdread(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
783787
int nbytes;
784788
MdfdVec *v;
785789

790+
/* If this build supports direct I/O, the buffer must be I/O aligned. */
791+
if (PG_O_DIRECT != 0 && PG_IO_ALIGN_SIZE <= BLCKSZ)
792+
Assert((uintptr_t) buffer == TYPEALIGN(PG_IO_ALIGN_SIZE, buffer));
793+
786794
TRACE_POSTGRESQL_SMGR_MD_READ_START(forknum, blocknum,
787795
reln->smgr_rlocator.locator.spcOid,
788796
reln->smgr_rlocator.locator.dbOid,
@@ -848,6 +856,10 @@ mdwrite(SMgrRelation reln, ForkNumber forknum, BlockNumber blocknum,
848856
int nbytes;
849857
MdfdVec *v;
850858

859+
/* If this build supports direct I/O, the buffer must be I/O aligned. */
860+
if (PG_O_DIRECT != 0 && PG_IO_ALIGN_SIZE <= BLCKSZ)
861+
Assert((uintptr_t) buffer == TYPEALIGN(PG_IO_ALIGN_SIZE, buffer));
862+
851863
/* This assert is too expensive to have on normally ... */
852864
#ifdef CHECK_WRITE_VS_EXTEND
853865
Assert(blocknum < mdnblocks(reln, forknum));
@@ -1429,7 +1441,8 @@ _mdfd_getseg(SMgrRelation reln, ForkNumber forknum, BlockNumber blkno,
14291441
*/
14301442
if (nblocks < ((BlockNumber) RELSEG_SIZE))
14311443
{
1432-
char *zerobuf = palloc0(BLCKSZ);
1444+
char *zerobuf = palloc_aligned(BLCKSZ, PG_IO_ALIGN_SIZE,
1445+
MCXT_ALLOC_ZERO);
14331446

14341447
mdextend(reln, forknum,
14351448
nextsegno * ((BlockNumber) RELSEG_SIZE) - 1,

‎src/backend/utils/sort/logtape.c

Copy file name to clipboardExpand all lines: src/backend/utils/sort/logtape.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -252,7 +252,7 @@ ltsWriteBlock(LogicalTapeSet *lts, long blocknum, const void *buffer)
252252
*/
253253
while (blocknum > lts->nBlocksWritten)
254254
{
255-
PGAlignedBlock zerobuf;
255+
PGIOAlignedBlock zerobuf;
256256

257257
MemSet(zerobuf.data, 0, sizeof(zerobuf));
258258

‎src/bin/pg_checksums/pg_checksums.c

Copy file name to clipboardExpand all lines: src/bin/pg_checksums/pg_checksums.c
+1-1Lines changed: 1 addition & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -183,7 +183,7 @@ skipfile(const char *fn)
183183
static void
184184
scan_file(const char *fn, int segmentno)
185185
{
186-
PGAlignedBlock buf;
186+
PGIOAlignedBlock buf;
187187
PageHeader header = (PageHeader) buf.data;
188188
int f;
189189
BlockNumber blockno;

‎src/bin/pg_rewind/local_source.c

Copy file name to clipboardExpand all lines: src/bin/pg_rewind/local_source.c
+2-2Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -77,7 +77,7 @@ static void
7777
local_queue_fetch_file(rewind_source *source, const char *path, size_t len)
7878
{
7979
const char *datadir = ((local_source *) source)->datadir;
80-
PGAlignedBlock buf;
80+
PGIOAlignedBlock buf;
8181
char srcpath[MAXPGPATH];
8282
int srcfd;
8383
size_t written_len;
@@ -129,7 +129,7 @@ local_queue_fetch_range(rewind_source *source, const char *path, off_t off,
129129
size_t len)
130130
{
131131
const char *datadir = ((local_source *) source)->datadir;
132-
PGAlignedBlock buf;
132+
PGIOAlignedBlock buf;
133133
char srcpath[MAXPGPATH];
134134
int srcfd;
135135
off_t begin = off;

‎src/bin/pg_upgrade/file.c

Copy file name to clipboardExpand all lines: src/bin/pg_upgrade/file.c
+2-2Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -178,8 +178,8 @@ rewriteVisibilityMap(const char *fromfile, const char *tofile,
178178
{
179179
int src_fd;
180180
int dst_fd;
181-
PGAlignedBlock buffer;
182-
PGAlignedBlock new_vmbuf;
181+
PGIOAlignedBlock buffer;
182+
PGIOAlignedBlock new_vmbuf;
183183
ssize_t totalBytesRead = 0;
184184
ssize_t src_filesize;
185185
int rewriteVmBytesPerPage;

‎src/common/file_utils.c

Copy file name to clipboardExpand all lines: src/common/file_utils.c
+2-2Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -540,8 +540,8 @@ pg_pwritev_with_retry(int fd, const struct iovec *iov, int iovcnt, off_t offset)
540540
ssize_t
541541
pg_pwrite_zeros(int fd, size_t size, off_t offset)
542542
{
543-
static const PGAlignedBlock zbuffer = {{0}}; /* worth BLCKSZ */
544-
void *zerobuf_addr = unconstify(PGAlignedBlock *, &zbuffer)->data;
543+
static const PGIOAlignedBlock zbuffer = {{0}}; /* worth BLCKSZ */
544+
void *zerobuf_addr = unconstify(PGIOAlignedBlock *, &zbuffer)->data;
545545
struct iovec iov[PG_IOV_MAX];
546546
size_t remaining_size = size;
547547
ssize_t total_written = 0;

‎src/include/c.h

Copy file name to clipboardExpand all lines: src/include/c.h
+26-8Lines changed: 26 additions & 8 deletions
Original file line numberDiff line numberDiff line change
@@ -1119,14 +1119,11 @@ extern void ExceptionalCondition(const char *conditionName,
11191119

11201120
/*
11211121
* Use this, not "char buf[BLCKSZ]", to declare a field or local variable
1122-
* holding a page buffer, if that page might be accessed as a page and not
1123-
* just a string of bytes. Otherwise the variable might be under-aligned,
1124-
* causing problems on alignment-picky hardware. (In some places, we use
1125-
* this to declare buffers even though we only pass them to read() and
1126-
* write(), because copying to/from aligned buffers is usually faster than
1127-
* using unaligned buffers.) We include both "double" and "int64" in the
1128-
* union to ensure that the compiler knows the value must be MAXALIGN'ed
1129-
* (cf. configure's computation of MAXIMUM_ALIGNOF).
1122+
* holding a page buffer, if that page might be accessed as a page. Otherwise
1123+
* the variable might be under-aligned, causing problems on alignment-picky
1124+
* hardware. We include both "double" and "int64" in the union to ensure that
1125+
* the compiler knows the value must be MAXALIGN'ed (cf. configure's
1126+
* computation of MAXIMUM_ALIGNOF).
11301127
*/
11311128
typedef union PGAlignedBlock
11321129
{
@@ -1135,9 +1132,30 @@ typedef union PGAlignedBlock
11351132
int64 force_align_i64;
11361133
} PGAlignedBlock;
11371134

1135+
/*
1136+
* Use this to declare a field or local variable holding a page buffer, if that
1137+
* page might be accessed as a page or passed to an SMgr I/O function. If
1138+
* allocating using the MemoryContext API, the aligned allocation functions
1139+
* should be used with PG_IO_ALIGN_SIZE. This alignment may be more efficient
1140+
* for I/O in general, but may be strictly required on some platforms when
1141+
* using direct I/O.
1142+
*/
1143+
typedef union PGIOAlignedBlock
1144+
{
1145+
#ifdef pg_attribute_aligned
1146+
pg_attribute_aligned(PG_IO_ALIGN_SIZE)
1147+
#endif
1148+
char data[BLCKSZ];
1149+
double force_align_d;
1150+
int64 force_align_i64;
1151+
} PGIOAlignedBlock;
1152+
11381153
/* Same, but for an XLOG_BLCKSZ-sized buffer */
11391154
typedef union PGAlignedXLogBlock
11401155
{
1156+
#ifdef pg_attribute_aligned
1157+
pg_attribute_aligned(PG_IO_ALIGN_SIZE)
1158+
#endif
11411159
char data[XLOG_BLCKSZ];
11421160
double force_align_d;
11431161
int64 force_align_i64;

0 commit comments

Comments
0 (0)
Morty Proxy This is a proxified and sanitized view of the page, visit original site.