| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* slab.c |
| 4 |
* SLAB allocator definitions. |
| 5 |
* |
| 6 |
* SLAB is a MemoryContext implementation designed for cases where large |
| 7 |
* numbers of equally-sized objects can be allocated and freed efficiently |
| 8 |
* with minimal memory wastage and fragmentation. |
| 9 |
* |
| 10 |
* |
| 11 |
* Portions Copyright (c) 2017-2025, PostgreSQL Global Development Group |
| 12 |
* |
| 13 |
* IDENTIFICATION |
| 14 |
* src/backend/utils/mmgr/slab.c |
| 15 |
* |
| 16 |
* |
| 17 |
* NOTE: |
| 18 |
* The constant allocation size allows significant simplification and various |
| 19 |
* optimizations over more general purpose allocators. The blocks are carved |
| 20 |
* into chunks of exactly the right size, wasting only the space required to |
| 21 |
* MAXALIGN the allocated chunks. |
| 22 |
* |
| 23 |
* Slab can also help reduce memory fragmentation in cases where longer-lived |
| 24 |
* chunks remain stored on blocks while most of the other chunks have already |
| 25 |
* been pfree'd. We give priority to putting new allocations into the |
| 26 |
* "fullest" block. This help avoid having too many sparsely used blocks |
| 27 |
* around and allows blocks to more easily become completely unused which |
| 28 |
* allows them to be eventually free'd. |
| 29 |
* |
| 30 |
* We identify the "fullest" block to put new allocations on by using a block |
| 31 |
* from the lowest populated element of the context's "blocklist" array. |
| 32 |
* This is an array of dlists containing blocks which we partition by the |
| 33 |
* number of free chunks which block has. Blocks with fewer free chunks are |
| 34 |
* stored in a lower indexed dlist array slot. Full blocks go on the 0th |
| 35 |
* element of the blocklist array. So that we don't have to have too many |
| 36 |
* elements in the array, each dlist in the array is responsible for a range |
| 37 |
* of free chunks. When a chunk is palloc'd or pfree'd we may need to move |
| 38 |
* the block onto another dlist if the number of free chunks crosses the |
| 39 |
* range boundary that the current list is responsible for. Having just a |
| 40 |
* few blocklist elements reduces the number of times we must move the block |
| 41 |
* onto another dlist element. |
| 42 |
* |
| 43 |
* We keep track of free chunks within each block by using a block-level free |
| 44 |
* list. We consult this list when we allocate a new chunk in the block. |
| 45 |
* The free list is a linked list, the head of which is pointed to with |
| 46 |
* SlabBlock's freehead field. Each subsequent list item is stored in the |
| 47 |
* free chunk's memory. We ensure chunks are large enough to store this |
| 48 |
* address. |
| 49 |
* |
| 50 |
* When we allocate a new block, technically all chunks are free, however, to |
| 51 |
* avoid having to write out the entire block to set the linked list for the |
| 52 |
* free chunks for every chunk in the block, we instead store a pointer to |
| 53 |
* the next "unused" chunk on the block and keep track of how many of these |
| 54 |
* unused chunks there are. When a new block is malloc'd, all chunks are |
| 55 |
* unused. The unused pointer starts with the first chunk on the block and |
| 56 |
* as chunks are allocated, the unused pointer is incremented. As chunks are |
| 57 |
* pfree'd, the unused pointer never goes backwards. The unused pointer can |
| 58 |
* be thought of as a high watermark for the maximum number of chunks in the |
| 59 |
* block which have been in use concurrently. When a chunk is pfree'd the |
| 60 |
* chunk is put onto the head of the free list and the unused pointer is not |
| 61 |
* changed. We only consume more unused chunks if we run out of free chunks |
| 62 |
* on the free list. This method effectively gives priority to using |
| 63 |
* previously used chunks over previously unused chunks, which should perform |
| 64 |
* better due to CPU caching effects. |
| 65 |
* |
| 66 |
*------------------------------------------------------------------------- |
| 67 |
*/ |
| 68 |
|
| 69 |
#include "postgres.h" |
| 70 |
|
| 71 |
#include "lib/ilist.h" |
| 72 |
#include "utils/memdebug.h" |
| 73 |
#include "utils/memutils.h" |
| 74 |
#include "utils/memutils_internal.h" |
| 75 |
#include "utils/memutils_memorychunk.h" |
| 76 |
|
| 77 |
#define Slab_BLOCKHDRSZ MAXALIGN(sizeof(SlabBlock)) |
| 78 |
|
| 79 |
#ifdef MEMORY_CONTEXT_CHECKING |
| 80 |
/* |
| 81 |
* Size of the memory required to store the SlabContext. |
| 82 |
* MEMORY_CONTEXT_CHECKING builds need some extra memory for the isChunkFree |
| 83 |
* array. |
| 84 |
*/ |
| 85 |
#define Slab_CONTEXT_HDRSZ(chunksPerBlock) \ |
| 86 |
(sizeof(SlabContext) + ((chunksPerBlock) * sizeof(bool))) |
| 87 |
#else |
| 88 |
#define Slab_CONTEXT_HDRSZ(chunksPerBlock) sizeof(SlabContext) |
| 89 |
#endif |
| 90 |
|
| 91 |
/* |
| 92 |
* The number of partitions to divide the blocklist into based their number of |
| 93 |
* free chunks. There must be at least 2. |
| 94 |
*/ |
| 95 |
#define SLAB_BLOCKLIST_COUNT 3 |
| 96 |
|
| 97 |
/* The maximum number of completely empty blocks to keep around for reuse. */ |
| 98 |
#define SLAB_MAXIMUM_EMPTY_BLOCKS 10 |
| 99 |
|
| 100 |
/* |
| 101 |
* SlabContext is a specialized implementation of MemoryContext. |
| 102 |
*/ |
| 103 |
typedef struct SlabContext |
| 104 |
{ |
| 105 |
MemoryContextData header; /* Standard memory-context fields */ |
| 106 |
/* Allocation parameters for this context: */ |
| 107 |
uint32 chunkSize; /* the requested (non-aligned) chunk size */ |
| 108 |
uint32 fullChunkSize; /* chunk size with chunk header and alignment */ |
| 109 |
uint32 blockSize; /* the size to make each block of chunks */ |
| 110 |
int32 chunksPerBlock; /* number of chunks that fit in 1 block */ |
| 111 |
int32 curBlocklistIndex; /* index into the blocklist[] element |
| 112 |
* containing the fullest, blocks */ |
| 113 |
#ifdef MEMORY_CONTEXT_CHECKING |
| 114 |
bool *isChunkFree; /* array to mark free chunks in a block during |
| 115 |
* SlabCheck */ |
| 116 |
#endif |
| 117 |
|
| 118 |
int32 blocklist_shift; /* number of bits to shift the nfree count |
| 119 |
* by to get the index into blocklist[] */ |
| 120 |
dclist_head emptyblocks; /* empty blocks to use up first instead of |
| 121 |
* mallocing new blocks */ |
| 122 |
|
| 123 |
/* |
| 124 |
* Blocks with free space, grouped by the number of free chunks they |
| 125 |
* contain. Completely full blocks are stored in the 0th element. |
| 126 |
* Completely empty blocks are stored in emptyblocks or free'd if we have |
| 127 |
* enough empty blocks already. |
| 128 |
*/ |
| 129 |
dlist_head blocklist[SLAB_BLOCKLIST_COUNT]; |
| 130 |
} SlabContext; |
| 131 |
|
| 132 |
/* |
| 133 |
* SlabBlock |
| 134 |
* Structure of a single slab block. |
| 135 |
* |
| 136 |
* slab: pointer back to the owning MemoryContext |
| 137 |
* nfree: number of chunks on the block which are unallocated |
| 138 |
* nunused: number of chunks on the block unallocated and not on the block's |
| 139 |
* freelist. |
| 140 |
* freehead: linked-list header storing a pointer to the first free chunk on |
| 141 |
* the block. Subsequent pointers are stored in the chunk's memory. NULL |
| 142 |
* indicates the end of the list. |
| 143 |
* unused: pointer to the next chunk which has yet to be used. |
| 144 |
* node: doubly-linked list node for the context's blocklist |
| 145 |
*/ |
| 146 |
typedef struct SlabBlock |
| 147 |
{ |
| 148 |
SlabContext *slab; /* owning context */ |
| 149 |
int32 nfree; /* number of chunks on free + unused chunks */ |
| 150 |
int32 nunused; /* number of unused chunks */ |
| 151 |
MemoryChunk *freehead; /* pointer to the first free chunk */ |
| 152 |
MemoryChunk *unused; /* pointer to the next unused chunk */ |
| 153 |
dlist_node node; /* doubly-linked list for blocklist[] */ |
| 154 |
} SlabBlock; |
| 155 |
|
| 156 |
|
| 157 |
#define Slab_CHUNKHDRSZ sizeof(MemoryChunk) |
| 158 |
#define SlabChunkGetPointer(chk) \ |
| 159 |
((void *) (((char *) (chk)) + sizeof(MemoryChunk))) |
| 160 |
|
| 161 |
/* |
| 162 |
* SlabBlockGetChunk |
| 163 |
* Obtain a pointer to the nth (0-based) chunk in the block |
| 164 |
*/ |
| 165 |
#define SlabBlockGetChunk(slab, block, n) \ |
| 166 |
((MemoryChunk *) ((char *) (block) + Slab_BLOCKHDRSZ \ |
| 167 |
+ ((n) * (slab)->fullChunkSize))) |
| 168 |
|
| 169 |
#if defined(MEMORY_CONTEXT_CHECKING) || defined(USE_ASSERT_CHECKING) |
| 170 |
|
| 171 |
/* |
| 172 |
* SlabChunkIndex |
| 173 |
* Get the 0-based index of how many chunks into the block the given |
| 174 |
* chunk is. |
| 175 |
*/ |
| 176 |
#define SlabChunkIndex(slab, block, chunk) \ |
| 177 |
(((char *) (chunk) - (char *) SlabBlockGetChunk(slab, block, 0)) / \ |
| 178 |
(slab)->fullChunkSize) |
| 179 |
|
| 180 |
/* |
| 181 |
* SlabChunkMod |
| 182 |
* A MemoryChunk should always be at an address which is a multiple of |
| 183 |
* fullChunkSize starting from the 0th chunk position. This will return |
| 184 |
* non-zero if it's not. |
| 185 |
*/ |
| 186 |
#define SlabChunkMod(slab, block, chunk) \ |
| 187 |
(((char *) (chunk) - (char *) SlabBlockGetChunk(slab, block, 0)) % \ |
| 188 |
(slab)->fullChunkSize) |
| 189 |
|
| 190 |
#endif |
| 191 |
|
| 192 |
/* |
| 193 |
* SlabIsValid |
| 194 |
* True iff set is a valid slab allocation set. |
| 195 |
*/ |
| 196 |
#define SlabIsValid(set) (PointerIsValid(set) && IsA(set, SlabContext)) |
| 197 |
|
| 198 |
/* |
| 199 |
* SlabBlockIsValid |
| 200 |
* True iff block is a valid block of slab allocation set. |
| 201 |
*/ |
| 202 |
#define SlabBlockIsValid(block) \ |
| 203 |
(PointerIsValid(block) && SlabIsValid((block)->slab)) |
| 204 |
|
| 205 |
/* |
| 206 |
* SlabBlocklistIndex |
| 207 |
* Determine the blocklist index that a block should be in for the given |
| 208 |
* number of free chunks. |
| 209 |
*/ |
| 210 |
static inline int32 |
| 211 |
SlabBlocklistIndex(SlabContext *slab, int nfree) |
| 212 |
{ |
| 213 |
int32 index; |
| 214 |
int32 blocklist_shift = slab->blocklist_shift; |
| 215 |
|
| 216 |
Assert(nfree >= 0 && nfree <= slab->chunksPerBlock); |
| 217 |
|
| 218 |
/* |
| 219 |
* Determine the blocklist index based on the number of free chunks. We |
| 220 |
* must ensure that 0 free chunks is dedicated to index 0. Everything |
| 221 |
* else must be >= 1 and < SLAB_BLOCKLIST_COUNT. |
| 222 |
* |
| 223 |
* To make this as efficient as possible, we exploit some two's complement |
| 224 |
* arithmetic where we reverse the sign before bit shifting. This results |
| 225 |
* in an nfree of 0 using index 0 and anything non-zero staying non-zero. |
| 226 |
* This is exploiting 0 and -0 being the same in two's complement. When |
| 227 |
* we're done, we just need to flip the sign back over again for a |
| 228 |
* positive index. |
| 229 |
*/ |
| 230 |
index = -((-nfree) >> blocklist_shift); |
| 231 |
|
| 232 |
if (nfree == 0) |
| 233 |
Assert(index == 0); |
| 234 |
else |
| 235 |
Assert(index >= 1 && index < SLAB_BLOCKLIST_COUNT); |
| 236 |
|
| 237 |
return index; |
| 238 |
} |
| 239 |
|
| 240 |
/* |
| 241 |
* SlabFindNextBlockListIndex |
| 242 |
* Search blocklist for blocks which have free chunks and return the |
| 243 |
* index of the blocklist found containing at least 1 block with free |
| 244 |
* chunks. If no block can be found we return 0. |
| 245 |
* |
| 246 |
* Note: We give priority to fuller blocks so that these are filled before |
| 247 |
* emptier blocks. This is done to increase the chances that mostly-empty |
| 248 |
* blocks will eventually become completely empty so they can be free'd. |
| 249 |
*/ |
| 250 |
static int32 |
| 251 |
SlabFindNextBlockListIndex(SlabContext *slab) |
| 252 |
{ |
| 253 |
/* start at 1 as blocklist[0] is for full blocks. */ |
| 254 |
for (int i = 1; i < SLAB_BLOCKLIST_COUNT; i++) |
| 255 |
{ |
| 256 |
/* return the first found non-empty index */ |
| 257 |
if (!dlist_is_empty(&slab->blocklist[i])) |
| 258 |
return i; |
| 259 |
} |
| 260 |
|
| 261 |
/* no blocks with free space */ |
| 262 |
return 0; |
| 263 |
} |
| 264 |
|
| 265 |
/* |
| 266 |
* SlabGetNextFreeChunk |
| 267 |
* Return the next free chunk in block and update the block to account |
| 268 |
* for the returned chunk now being used. |
| 269 |
*/ |
| 270 |
static inline MemoryChunk * |
| 271 |
SlabGetNextFreeChunk(SlabContext *slab, SlabBlock *block) |
| 272 |
{ |
| 273 |
MemoryChunk *chunk; |
| 274 |
|
| 275 |
Assert(block->nfree > 0); |
| 276 |
|
| 277 |
if (block->freehead != NULL) |
| 278 |
{ |
| 279 |
chunk = block->freehead; |
| 280 |
|
| 281 |
/* |
| 282 |
* Pop the chunk from the linked list of free chunks. The pointer to |
| 283 |
* the next free chunk is stored in the chunk itself. |
| 284 |
*/ |
| 285 |
VALGRIND_MAKE_MEM_DEFINED(SlabChunkGetPointer(chunk), sizeof(MemoryChunk *)); |
| 286 |
block->freehead = *(MemoryChunk **) SlabChunkGetPointer(chunk); |
| 287 |
|
| 288 |
/* check nothing stomped on the free chunk's memory */ |
| 289 |
Assert(block->freehead == NULL || |
| 290 |
(block->freehead >= SlabBlockGetChunk(slab, block, 0) && |
| 291 |
block->freehead <= SlabBlockGetChunk(slab, block, slab->chunksPerBlock - 1) && |
| 292 |
SlabChunkMod(slab, block, block->freehead) == 0)); |
| 293 |
} |
| 294 |
else |
| 295 |
{ |
| 296 |
Assert(block->nunused > 0); |
| 297 |
|
| 298 |
chunk = block->unused; |
| 299 |
block->unused = (MemoryChunk *) (((char *) block->unused) + slab->fullChunkSize); |
| 300 |
block->nunused--; |
| 301 |
} |
| 302 |
|
| 303 |
block->nfree--; |
| 304 |
|
| 305 |
return chunk; |
| 306 |
} |
| 307 |
|
| 308 |
/* |
| 309 |
* SlabContextCreate |
| 310 |
* Create a new Slab context. |
| 311 |
* |
| 312 |
* parent: parent context, or NULL if top-level context |
| 313 |
* name: name of context (must be statically allocated) |
| 314 |
* blockSize: allocation block size |
| 315 |
* chunkSize: allocation chunk size |
| 316 |
* |
| 317 |
* The Slab_CHUNKHDRSZ + MAXALIGN(chunkSize + 1) may not exceed |
| 318 |
* MEMORYCHUNK_MAX_VALUE. |
| 319 |
* 'blockSize' may not exceed MEMORYCHUNK_MAX_BLOCKOFFSET. |
| 320 |
*/ |
| 321 |
MemoryContext |
| 322 |
SlabContextCreate(MemoryContext parent, |
| 323 |
const char *name, |
| 324 |
Size blockSize, |
| 325 |
Size chunkSize) |
| 326 |
{ |
| 327 |
int chunksPerBlock; |
| 328 |
Size fullChunkSize; |
| 329 |
SlabContext *slab; |
| 330 |
int i; |
| 331 |
|
| 332 |
/* ensure MemoryChunk's size is properly maxaligned */ |
| 333 |
StaticAssertDecl(Slab_CHUNKHDRSZ == MAXALIGN(Slab_CHUNKHDRSZ), |
| 334 |
"sizeof(MemoryChunk) is not maxaligned"); |
| 335 |
Assert(blockSize <= MEMORYCHUNK_MAX_BLOCKOFFSET); |
| 336 |
|
| 337 |
/* |
| 338 |
* Ensure there's enough space to store the pointer to the next free chunk |
| 339 |
* in the memory of the (otherwise) unused allocation. |
| 340 |
*/ |
| 341 |
if (chunkSize < sizeof(MemoryChunk *)) |
| 342 |
chunkSize = sizeof(MemoryChunk *); |
| 343 |
|
| 344 |
/* length of the maxaligned chunk including the chunk header */ |
| 345 |
#ifdef MEMORY_CONTEXT_CHECKING |
| 346 |
/* ensure there's always space for the sentinel byte */ |
| 347 |
fullChunkSize = Slab_CHUNKHDRSZ + MAXALIGN(chunkSize + 1); |
| 348 |
#else |
| 349 |
fullChunkSize = Slab_CHUNKHDRSZ + MAXALIGN(chunkSize); |
| 350 |
#endif |
| 351 |
|
| 352 |
Assert(fullChunkSize <= MEMORYCHUNK_MAX_VALUE); |
| 353 |
|
| 354 |
/* compute the number of chunks that will fit on each block */ |
| 355 |
chunksPerBlock = (blockSize - Slab_BLOCKHDRSZ) / fullChunkSize; |
| 356 |
|
| 357 |
/* Make sure the block can store at least one chunk. */ |
| 358 |
if (chunksPerBlock == 0) |
| 359 |
elog(ERROR, "block size %zu for slab is too small for %zu-byte chunks", |
| 360 |
blockSize, chunkSize); |
| 361 |
|
| 362 |
|
| 363 |
|
| 364 |
slab = (SlabContext *) malloc(Slab_CONTEXT_HDRSZ(chunksPerBlock)); |
| 365 |
if (slab == NULL) |
| 366 |
{ |
| 367 |
MemoryContextStats(TopMemoryContext); |
| 368 |
ereport(ERROR, |
| 369 |
(errcode(ERRCODE_OUT_OF_MEMORY), |
| 370 |
errmsg("out of memory"), |
| 371 |
errdetail("Failed while creating memory context \"%s\".", |
| 372 |
name))); |
| 373 |
} |
| 374 |
|
| 375 |
/* |
| 376 |
* Avoid writing code that can fail between here and MemoryContextCreate; |
| 377 |
* we'd leak the header if we ereport in this stretch. |
| 378 |
*/ |
| 379 |
|
| 380 |
/* Fill in SlabContext-specific header fields */ |
| 381 |
slab->chunkSize = (uint32) chunkSize; |
| 382 |
slab->fullChunkSize = (uint32) fullChunkSize; |
| 383 |
slab->blockSize = (uint32) blockSize; |
| 384 |
slab->chunksPerBlock = chunksPerBlock; |
| 385 |
slab->curBlocklistIndex = 0; |
| 386 |
|
| 387 |
/* |
| 388 |
* Compute a shift that guarantees that shifting chunksPerBlock with it is |
| 389 |
* < SLAB_BLOCKLIST_COUNT - 1. The reason that we subtract 1 from |
| 390 |
* SLAB_BLOCKLIST_COUNT in this calculation is that we reserve the 0th |
| 391 |
* blocklist element for blocks which have no free chunks. |
| 392 |
* |
| 393 |
* We calculate the number of bits to shift by rather than a divisor to |
| 394 |
* divide by as performing division each time we need to find the |
| 395 |
* blocklist index would be much slower. |
| 396 |
*/ |
| 397 |
slab->blocklist_shift = 0; |
| 398 |
while ((slab->chunksPerBlock >> slab->blocklist_shift) >= (SLAB_BLOCKLIST_COUNT - 1)) |
| 399 |
slab->blocklist_shift++; |
| 400 |
|
| 401 |
/* initialize the list to store empty blocks to be reused */ |
| 402 |
dclist_init(&slab->emptyblocks); |
| 403 |
|
| 404 |
/* initialize each blocklist slot */ |
| 405 |
for (i = 0; i < SLAB_BLOCKLIST_COUNT; i++) |
| 406 |
dlist_init(&slab->blocklist[i]); |
| 407 |
|
| 408 |
#ifdef MEMORY_CONTEXT_CHECKING |
| 409 |
/* set the isChunkFree pointer right after the end of the context */ |
| 410 |
slab->isChunkFree = (bool *) ((char *) slab + sizeof(SlabContext)); |
| 411 |
#endif |
| 412 |
|
| 413 |
/* Finally, do the type-independent part of context creation */ |
| 414 |
MemoryContextCreate((MemoryContext) slab, |
| 415 |
T_SlabContext, |
| 416 |
MCTX_SLAB_ID, |
| 417 |
parent, |
| 418 |
name); |
| 419 |
|
| 420 |
return (MemoryContext) slab; |
| 421 |
} |
| 422 |
|
| 423 |
/* |
| 424 |
* SlabReset |
| 425 |
* Frees all memory which is allocated in the given set. |
| 426 |
* |
| 427 |
* The code simply frees all the blocks in the context - we don't keep any |
| 428 |
* keeper blocks or anything like that. |
| 429 |
*/ |
| 430 |
void |
| 431 |
SlabReset(MemoryContext context) |
| 432 |
{ |
| 433 |
SlabContext *slab = (SlabContext *) context; |
| 434 |
dlist_mutable_iter miter; |
| 435 |
int i; |
| 436 |
|
| 437 |
Assert(SlabIsValid(slab)); |
| 438 |
|
| 439 |
#ifdef MEMORY_CONTEXT_CHECKING |
| 440 |
/* Check for corruption and leaks before freeing */ |
| 441 |
SlabCheck(context); |
| 442 |
#endif |
| 443 |
|
| 444 |
/* release any retained empty blocks */ |
| 445 |
dclist_foreach_modify(miter, &slab->emptyblocks) |
| 446 |
{ |
| 447 |
SlabBlock *block = dlist_container(SlabBlock, node, miter.cur); |
| 448 |
|
| 449 |
dclist_delete_from(&slab->emptyblocks, miter.cur); |
| 450 |
|
| 451 |
#ifdef CLOBBER_FREED_MEMORY |
| 452 |
wipe_mem(block, slab->blockSize); |
| 453 |
#endif |
| 454 |
free(block); |
| 455 |
context->mem_allocated -= slab->blockSize; |
| 456 |
} |
| 457 |
|
| 458 |
/* walk over blocklist and free the blocks */ |
| 459 |
for (i = 0; i < SLAB_BLOCKLIST_COUNT; i++) |
| 460 |
{ |
| 461 |
dlist_foreach_modify(miter, &slab->blocklist[i]) |
| 462 |
{ |
| 463 |
SlabBlock *block = dlist_container(SlabBlock, node, miter.cur); |
| 464 |
|
| 465 |
dlist_delete(miter.cur); |
| 466 |
|
| 467 |
#ifdef CLOBBER_FREED_MEMORY |
| 468 |
wipe_mem(block, slab->blockSize); |
| 469 |
#endif |
| 470 |
free(block); |
| 471 |
context->mem_allocated -= slab->blockSize; |
| 472 |
} |
| 473 |
} |
| 474 |
|
| 475 |
slab->curBlocklistIndex = 0; |
| 476 |
|
| 477 |
Assert(context->mem_allocated == 0); |
| 478 |
} |
| 479 |
|
| 480 |
/* |
| 481 |
* SlabDelete |
| 482 |
* Free all memory which is allocated in the given context. |
| 483 |
*/ |
| 484 |
void |
| 485 |
SlabDelete(MemoryContext context) |
| 486 |
{ |
| 487 |
/* Reset to release all the SlabBlocks */ |
| 488 |
SlabReset(context); |
| 489 |
/* And free the context header */ |
| 490 |
free(context); |
| 491 |
} |
| 492 |
|
| 493 |
/* |
| 494 |
* Small helper for allocating a new chunk from a chunk, to avoid duplicating |
| 495 |
* the code between SlabAlloc() and SlabAllocFromNewBlock(). |
| 496 |
*/ |
| 497 |
static inline void * |
| 498 |
SlabAllocSetupNewChunk(MemoryContext context, SlabBlock *block, |
| 499 |
MemoryChunk *chunk, Size size) |
| 500 |
{ |
| 501 |
SlabContext *slab = (SlabContext *) context; |
| 502 |
|
| 503 |
/* |
| 504 |
* Check that the chunk pointer is actually somewhere on the block and is |
| 505 |
* aligned as expected. |
| 506 |
*/ |
| 507 |
Assert(chunk >= SlabBlockGetChunk(slab, block, 0)); |
| 508 |
Assert(chunk <= SlabBlockGetChunk(slab, block, slab->chunksPerBlock - 1)); |
| 509 |
Assert(SlabChunkMod(slab, block, chunk) == 0); |
| 510 |
|
| 511 |
/* Prepare to initialize the chunk header. */ |
| 512 |
VALGRIND_MAKE_MEM_UNDEFINED(chunk, Slab_CHUNKHDRSZ); |
| 513 |
|
| 514 |
MemoryChunkSetHdrMask(chunk, block, MAXALIGN(slab->chunkSize), MCTX_SLAB_ID); |
| 515 |
|
| 516 |
#ifdef MEMORY_CONTEXT_CHECKING |
| 517 |
chunk->requested_size = size; |
| 518 |
/* slab mark to catch clobber of "unused" space */ |
| 519 |
Assert(slab->chunkSize < (slab->fullChunkSize - Slab_CHUNKHDRSZ)); |
| 520 |
set_sentinel(MemoryChunkGetPointer(chunk), size); |
| 521 |
VALGRIND_MAKE_MEM_NOACCESS(((char *) chunk) + Slab_CHUNKHDRSZ + |
| 522 |
slab->chunkSize, |
| 523 |
slab->fullChunkSize - |
| 524 |
(slab->chunkSize + Slab_CHUNKHDRSZ)); |
| 525 |
#endif |
| 526 |
|
| 527 |
#ifdef RANDOMIZE_ALLOCATED_MEMORY |
| 528 |
/* fill the allocated space with junk */ |
| 529 |
randomize_mem((char *) MemoryChunkGetPointer(chunk), size); |
| 530 |
#endif |
| 531 |
|
| 532 |
/* Disallow access to the chunk header. */ |
| 533 |
VALGRIND_MAKE_MEM_NOACCESS(chunk, Slab_CHUNKHDRSZ); |
| 534 |
|
| 535 |
return MemoryChunkGetPointer(chunk); |
| 536 |
} |
| 537 |
|
| 538 |
pg_noinline |
| 539 |
static void * |
| 540 |
SlabAllocFromNewBlock(MemoryContext context, Size size, int flags) |
| 541 |
{ |
| 542 |
SlabContext *slab = (SlabContext *) context; |
| 543 |
SlabBlock *block; |
| 544 |
MemoryChunk *chunk; |
| 545 |
dlist_head *blocklist; |
| 546 |
int blocklist_idx; |
| 547 |
|
| 548 |
/* to save allocating a new one, first check the empty blocks list */ |
| 549 |
if (dclist_count(&slab->emptyblocks) > 0) |
| 550 |
{ |
| 551 |
dlist_node *node = dclist_pop_head_node(&slab->emptyblocks); |
| 552 |
|
| 553 |
block = dlist_container(SlabBlock, node, node); |
| 554 |
|
| 555 |
/* |
| 556 |
* SlabFree() should have left this block in a valid state with all |
| 557 |
* chunks free. Ensure that's the case. |
| 558 |
*/ |
| 559 |
Assert(block->nfree == slab->chunksPerBlock); |
| 560 |
|
| 561 |
/* fetch the next chunk from this block */ |
| 562 |
chunk = SlabGetNextFreeChunk(slab, block); |
| 563 |
} |
| 564 |
else |
| 565 |
{ |
| 566 |
block = (SlabBlock *) malloc(slab->blockSize); |
| 567 |
|
| 568 |
if (unlikely(block == NULL)) |
| 569 |
return MemoryContextAllocationFailure(context, size, flags); |
| 570 |
|
| 571 |
block->slab = slab; |
| 572 |
context->mem_allocated += slab->blockSize; |
| 573 |
|
| 574 |
/* use the first chunk in the new block */ |
| 575 |
chunk = SlabBlockGetChunk(slab, block, 0); |
| 576 |
|
| 577 |
block->nfree = slab->chunksPerBlock - 1; |
| 578 |
block->unused = SlabBlockGetChunk(slab, block, 1); |
| 579 |
block->freehead = NULL; |
| 580 |
block->nunused = slab->chunksPerBlock - 1; |
| 581 |
} |
| 582 |
|
| 583 |
/* find the blocklist element for storing blocks with 1 used chunk */ |
| 584 |
blocklist_idx = SlabBlocklistIndex(slab, block->nfree); |
| 585 |
blocklist = &slab->blocklist[blocklist_idx]; |
| 586 |
|
| 587 |
/* this better be empty. We just added a block thinking it was */ |
| 588 |
Assert(dlist_is_empty(blocklist)); |
| 589 |
|
| 590 |
dlist_push_head(blocklist, &block->node); |
| 591 |
|
| 592 |
slab->curBlocklistIndex = blocklist_idx; |
| 593 |
|
| 594 |
return SlabAllocSetupNewChunk(context, block, chunk, size); |
| 595 |
} |
| 596 |
|
| 597 |
/* |
| 598 |
* SlabAllocInvalidSize |
| 599 |
* Handle raising an ERROR for an invalid size request. We don't do this |
| 600 |
* in slab alloc as calling the elog functions would force the compiler |
| 601 |
* to setup the stack frame in SlabAlloc. For performance reasons, we |
| 602 |
* want to avoid that. |
| 603 |
*/ |
| 604 |
pg_noinline |
| 605 |
pg_noreturn |
| 606 |
static void |
| 607 |
SlabAllocInvalidSize(MemoryContext context, Size size) |
| 608 |
{ |
| 609 |
SlabContext *slab = (SlabContext *) context; |
| 610 |
|
| 611 |
elog(ERROR, "unexpected alloc chunk size %zu (expected %u)", size, |
| 612 |
slab->chunkSize); |
| 613 |
} |
| 614 |
|
| 615 |
/* |
| 616 |
* SlabAlloc |
| 617 |
* Returns a pointer to a newly allocated memory chunk or raises an ERROR |
| 618 |
* on allocation failure, or returns NULL when flags contains |
| 619 |
* MCXT_ALLOC_NO_OOM. 'size' must be the same size as was specified |
| 620 |
* during SlabContextCreate(). |
| 621 |
* |
| 622 |
* This function should only contain the most common code paths. Everything |
| 623 |
* else should be in pg_noinline helper functions, thus avoiding the overhead |
| 624 |
* of creating a stack frame for the common cases. Allocating memory is often |
| 625 |
* a bottleneck in many workloads, so avoiding stack frame setup is |
| 626 |
* worthwhile. Helper functions should always directly return the newly |
| 627 |
* allocated memory so that we can just return that address directly as a tail |
| 628 |
* call. |
| 629 |
*/ |
| 630 |
void * |
| 631 |
SlabAlloc(MemoryContext context, Size size, int flags) |
| 632 |
{ |
| 633 |
SlabContext *slab = (SlabContext *) context; |
| 634 |
SlabBlock *block; |
| 635 |
MemoryChunk *chunk; |
| 636 |
|
| 637 |
Assert(SlabIsValid(slab)); |
| 638 |
|
| 639 |
/* sanity check that this is pointing to a valid blocklist */ |
| 640 |
Assert(slab->curBlocklistIndex >= 0); |
| 641 |
Assert(slab->curBlocklistIndex <= SlabBlocklistIndex(slab, slab->chunksPerBlock)); |
| 642 |
|
| 643 |
/* |
| 644 |
* Make sure we only allow correct request size. This doubles as the |
| 645 |
* MemoryContextCheckSize check. |
| 646 |
*/ |
| 647 |
if (unlikely(size != slab->chunkSize)) |
| 648 |
SlabAllocInvalidSize(context, size); |
| 649 |
|
| 650 |
if (unlikely(slab->curBlocklistIndex == 0)) |
| 651 |
{ |
| 652 |
/* |
| 653 |
* Handle the case when there are no partially filled blocks |
| 654 |
* available. This happens either when the last allocation took the |
| 655 |
* last chunk in the block, or when SlabFree() free'd the final block. |
| 656 |
*/ |
| 657 |
return SlabAllocFromNewBlock(context, size, flags); |
| 658 |
} |
| 659 |
else |
| 660 |
{ |
| 661 |
dlist_head *blocklist = &slab->blocklist[slab->curBlocklistIndex]; |
| 662 |
int new_blocklist_idx; |
| 663 |
|
| 664 |
Assert(!dlist_is_empty(blocklist)); |
| 665 |
|
| 666 |
/* grab the block from the blocklist */ |
| 667 |
block = dlist_head_element(SlabBlock, node, blocklist); |
| 668 |
|
| 669 |
/* make sure we actually got a valid block, with matching nfree */ |
| 670 |
Assert(block != NULL); |
| 671 |
Assert(slab->curBlocklistIndex == SlabBlocklistIndex(slab, block->nfree)); |
| 672 |
Assert(block->nfree > 0); |
| 673 |
|
| 674 |
/* fetch the next chunk from this block */ |
| 675 |
chunk = SlabGetNextFreeChunk(slab, block); |
| 676 |
|
| 677 |
/* get the new blocklist index based on the new free chunk count */ |
| 678 |
new_blocklist_idx = SlabBlocklistIndex(slab, block->nfree); |
| 679 |
|
| 680 |
/* |
| 681 |
* Handle the case where the blocklist index changes. This also deals |
| 682 |
* with blocks becoming full as only full blocks go at index 0. |
| 683 |
*/ |
| 684 |
if (unlikely(slab->curBlocklistIndex != new_blocklist_idx)) |
| 685 |
{ |
| 686 |
dlist_delete_from(blocklist, &block->node); |
| 687 |
dlist_push_head(&slab->blocklist[new_blocklist_idx], &block->node); |
| 688 |
|
| 689 |
if (dlist_is_empty(blocklist)) |
| 690 |
slab->curBlocklistIndex = SlabFindNextBlockListIndex(slab); |
| 691 |
} |
| 692 |
} |
| 693 |
|
| 694 |
return SlabAllocSetupNewChunk(context, block, chunk, size); |
| 695 |
} |
| 696 |
|
| 697 |
/* |
| 698 |
* SlabFree |
| 699 |
* Frees allocated memory; memory is removed from the slab. |
| 700 |
*/ |
| 701 |
void |
| 702 |
SlabFree(void *pointer) |
| 703 |
{ |
| 704 |
MemoryChunk *chunk = PointerGetMemoryChunk(pointer); |
| 705 |
SlabBlock *block; |
| 706 |
SlabContext *slab; |
| 707 |
int curBlocklistIdx; |
| 708 |
int newBlocklistIdx; |
| 709 |
|
| 710 |
/* Allow access to the chunk header. */ |
| 711 |
VALGRIND_MAKE_MEM_DEFINED(chunk, Slab_CHUNKHDRSZ); |
| 712 |
|
| 713 |
block = MemoryChunkGetBlock(chunk); |
| 714 |
|
| 715 |
/* |
| 716 |
* For speed reasons we just Assert that the referenced block is good. |
| 717 |
* Future field experience may show that this Assert had better become a |
| 718 |
* regular runtime test-and-elog check. |
| 719 |
*/ |
| 720 |
Assert(SlabBlockIsValid(block)); |
| 721 |
slab = block->slab; |
| 722 |
|
| 723 |
#ifdef MEMORY_CONTEXT_CHECKING |
| 724 |
/* See comments in AllocSetFree about uses of ERROR and WARNING here */ |
| 725 |
/* Test for previously-freed chunk */ |
| 726 |
if (unlikely(chunk->requested_size == InvalidAllocSize)) |
| 727 |
elog(ERROR, "detected double pfree in %s %p", |
| 728 |
slab->header.name, chunk); |
| 729 |
/* Test for someone scribbling on unused space in chunk */ |
| 730 |
Assert(slab->chunkSize < (slab->fullChunkSize - Slab_CHUNKHDRSZ)); |
| 731 |
if (!sentinel_ok(pointer, slab->chunkSize)) |
| 732 |
elog(WARNING, "detected write past chunk end in %s %p", |
| 733 |
slab->header.name, chunk); |
| 734 |
/* Reset requested_size to InvalidAllocSize in free chunks */ |
| 735 |
chunk->requested_size = InvalidAllocSize; |
| 736 |
#endif |
| 737 |
|
| 738 |
/* push this chunk onto the head of the block's free list */ |
| 739 |
*(MemoryChunk **) pointer = block->freehead; |
| 740 |
block->freehead = chunk; |
| 741 |
|
| 742 |
block->nfree++; |
| 743 |
|
| 744 |
Assert(block->nfree > 0); |
| 745 |
Assert(block->nfree <= slab->chunksPerBlock); |
| 746 |
|
| 747 |
#ifdef CLOBBER_FREED_MEMORY |
| 748 |
/* don't wipe the free list MemoryChunk pointer stored in the chunk */ |
| 749 |
wipe_mem((char *) pointer + sizeof(MemoryChunk *), |
| 750 |
slab->chunkSize - sizeof(MemoryChunk *)); |
| 751 |
#endif |
| 752 |
|
| 753 |
curBlocklistIdx = SlabBlocklistIndex(slab, block->nfree - 1); |
| 754 |
newBlocklistIdx = SlabBlocklistIndex(slab, block->nfree); |
| 755 |
|
| 756 |
/* |
| 757 |
* Check if the block needs to be moved to another element on the |
| 758 |
* blocklist based on it now having 1 more free chunk. |
| 759 |
*/ |
| 760 |
if (unlikely(curBlocklistIdx != newBlocklistIdx)) |
| 761 |
{ |
| 762 |
/* do the move */ |
| 763 |
dlist_delete_from(&slab->blocklist[curBlocklistIdx], &block->node); |
| 764 |
dlist_push_head(&slab->blocklist[newBlocklistIdx], &block->node); |
| 765 |
|
| 766 |
/* |
| 767 |
* The blocklist[curBlocklistIdx] may now be empty or we may now be |
| 768 |
* able to use a lower-element blocklist. We'll need to redetermine |
| 769 |
* what the slab->curBlocklistIndex is if the current blocklist was |
| 770 |
* changed or if a lower element one was changed. We must ensure we |
| 771 |
* use the list with the fullest block(s). |
| 772 |
*/ |
| 773 |
if (slab->curBlocklistIndex >= curBlocklistIdx) |
| 774 |
{ |
| 775 |
slab->curBlocklistIndex = SlabFindNextBlockListIndex(slab); |
| 776 |
|
| 777 |
/* |
| 778 |
* We know there must be a block with at least 1 unused chunk as |
| 779 |
* we just pfree'd one. Ensure curBlocklistIndex reflects this. |
| 780 |
*/ |
| 781 |
Assert(slab->curBlocklistIndex > 0); |
| 782 |
} |
| 783 |
} |
| 784 |
|
| 785 |
/* Handle when a block becomes completely empty */ |
| 786 |
if (unlikely(block->nfree == slab->chunksPerBlock)) |
| 787 |
{ |
| 788 |
/* remove the block */ |
| 789 |
dlist_delete_from(&slab->blocklist[newBlocklistIdx], &block->node); |
| 790 |
|
| 791 |
/* |
| 792 |
* To avoid thrashing malloc/free, we keep a list of empty blocks that |
| 793 |
* we can reuse again instead of having to malloc a new one. |
| 794 |
*/ |
| 795 |
if (dclist_count(&slab->emptyblocks) < SLAB_MAXIMUM_EMPTY_BLOCKS) |
| 796 |
dclist_push_head(&slab->emptyblocks, &block->node); |
| 797 |
else |
| 798 |
{ |
| 799 |
/* |
| 800 |
* When we have enough empty blocks stored already, we actually |
| 801 |
* free the block. |
| 802 |
*/ |
| 803 |
#ifdef CLOBBER_FREED_MEMORY |
| 804 |
wipe_mem(block, slab->blockSize); |
| 805 |
#endif |
| 806 |
free(block); |
| 807 |
slab->header.mem_allocated -= slab->blockSize; |
| 808 |
} |
| 809 |
|
| 810 |
/* |
| 811 |
* Check if we need to reset the blocklist index. This is required |
| 812 |
* when the blocklist this block is on has become completely empty. |
| 813 |
*/ |
| 814 |
if (slab->curBlocklistIndex == newBlocklistIdx && |
| 815 |
dlist_is_empty(&slab->blocklist[newBlocklistIdx])) |
| 816 |
slab->curBlocklistIndex = SlabFindNextBlockListIndex(slab); |
| 817 |
} |
| 818 |
} |
| 819 |
|
| 820 |
/* |
| 821 |
* SlabRealloc |
| 822 |
* Change the allocated size of a chunk. |
| 823 |
* |
| 824 |
* As Slab is designed for allocating equally-sized chunks of memory, it can't |
| 825 |
* do an actual chunk size change. We try to be gentle and allow calls with |
| 826 |
* exactly the same size, as in that case we can simply return the same |
| 827 |
* chunk. When the size differs, we throw an error. |
| 828 |
* |
| 829 |
* We could also allow requests with size < chunkSize. That however seems |
| 830 |
* rather pointless - Slab is meant for chunks of constant size, and moreover |
| 831 |
* realloc is usually used to enlarge the chunk. |
| 832 |
*/ |
| 833 |
void * |
| 834 |
SlabRealloc(void *pointer, Size size, int flags) |
| 835 |
{ |
| 836 |
MemoryChunk *chunk = PointerGetMemoryChunk(pointer); |
| 837 |
SlabBlock *block; |
| 838 |
SlabContext *slab; |
| 839 |
|
| 840 |
/* Allow access to the chunk header. */ |
| 841 |
VALGRIND_MAKE_MEM_DEFINED(chunk, Slab_CHUNKHDRSZ); |
| 842 |
|
| 843 |
block = MemoryChunkGetBlock(chunk); |
| 844 |
|
| 845 |
/* Disallow access to the chunk header. */ |
| 846 |
VALGRIND_MAKE_MEM_NOACCESS(chunk, Slab_CHUNKHDRSZ); |
| 847 |
|
| 848 |
/* |
| 849 |
* Try to verify that we have a sane block pointer: the block header |
| 850 |
* should reference a slab context. (We use a test-and-elog, not just |
| 851 |
* Assert, because it seems highly likely that we're here in error in the |
| 852 |
* first place.) |
| 853 |
*/ |
| 854 |
if (!SlabBlockIsValid(block)) |
| 855 |
elog(ERROR, "could not find block containing chunk %p", chunk); |
| 856 |
slab = block->slab; |
| 857 |
|
| 858 |
/* can't do actual realloc with slab, but let's try to be gentle */ |
| 859 |
if (size == slab->chunkSize) |
| 860 |
return pointer; |
| 861 |
|
| 862 |
elog(ERROR, "slab allocator does not support realloc()"); |
| 863 |
return NULL; /* keep compiler quiet */ |
| 864 |
} |
| 865 |
|
| 866 |
/* |
| 867 |
* SlabGetChunkContext |
| 868 |
* Return the MemoryContext that 'pointer' belongs to. |
| 869 |
*/ |
| 870 |
MemoryContext |
| 871 |
SlabGetChunkContext(void *pointer) |
| 872 |
{ |
| 873 |
MemoryChunk *chunk = PointerGetMemoryChunk(pointer); |
| 874 |
SlabBlock *block; |
| 875 |
|
| 876 |
/* Allow access to the chunk header. */ |
| 877 |
VALGRIND_MAKE_MEM_DEFINED(chunk, Slab_CHUNKHDRSZ); |
| 878 |
|
| 879 |
block = MemoryChunkGetBlock(chunk); |
| 880 |
|
| 881 |
/* Disallow access to the chunk header. */ |
| 882 |
VALGRIND_MAKE_MEM_NOACCESS(chunk, Slab_CHUNKHDRSZ); |
| 883 |
|
| 884 |
Assert(SlabBlockIsValid(block)); |
| 885 |
|
| 886 |
return &block->slab->header; |
| 887 |
} |
| 888 |
|
| 889 |
/* |
| 890 |
* SlabGetChunkSpace |
| 891 |
* Given a currently-allocated chunk, determine the total space |
| 892 |
* it occupies (including all memory-allocation overhead). |
| 893 |
*/ |
| 894 |
Size |
| 895 |
SlabGetChunkSpace(void *pointer) |
| 896 |
{ |
| 897 |
MemoryChunk *chunk = PointerGetMemoryChunk(pointer); |
| 898 |
SlabBlock *block; |
| 899 |
SlabContext *slab; |
| 900 |
|
| 901 |
/* Allow access to the chunk header. */ |
| 902 |
VALGRIND_MAKE_MEM_DEFINED(chunk, Slab_CHUNKHDRSZ); |
| 903 |
|
| 904 |
block = MemoryChunkGetBlock(chunk); |
| 905 |
|
| 906 |
/* Disallow access to the chunk header. */ |
| 907 |
VALGRIND_MAKE_MEM_NOACCESS(chunk, Slab_CHUNKHDRSZ); |
| 908 |
|
| 909 |
Assert(SlabBlockIsValid(block)); |
| 910 |
slab = block->slab; |
| 911 |
|
| 912 |
return slab->fullChunkSize; |
| 913 |
} |
| 914 |
|
| 915 |
/* |
| 916 |
* SlabIsEmpty |
| 917 |
* Is the slab empty of any allocated space? |
| 918 |
*/ |
| 919 |
bool |
| 920 |
SlabIsEmpty(MemoryContext context) |
| 921 |
{ |
| 922 |
Assert(SlabIsValid((SlabContext *) context)); |
| 923 |
|
| 924 |
return (context->mem_allocated == 0); |
| 925 |
} |
| 926 |
|
| 927 |
/* |
| 928 |
* SlabStats |
| 929 |
* Compute stats about memory consumption of a Slab context. |
| 930 |
* |
| 931 |
* printfunc: if not NULL, pass a human-readable stats string to this. |
| 932 |
* passthru: pass this pointer through to printfunc. |
| 933 |
* totals: if not NULL, add stats about this context into *totals. |
| 934 |
* print_to_stderr: print stats to stderr if true, elog otherwise. |
| 935 |
*/ |
| 936 |
void |
| 937 |
SlabStats(MemoryContext context, |
| 938 |
MemoryStatsPrintFunc printfunc, void *passthru, |
| 939 |
MemoryContextCounters *totals, |
| 940 |
bool print_to_stderr) |
| 941 |
{ |
| 942 |
SlabContext *slab = (SlabContext *) context; |
| 943 |
Size nblocks = 0; |
| 944 |
Size freechunks = 0; |
| 945 |
Size totalspace; |
| 946 |
Size freespace = 0; |
| 947 |
int i; |
| 948 |
|
| 949 |
Assert(SlabIsValid(slab)); |
| 950 |
|
| 951 |
/* Include context header in totalspace */ |
| 952 |
totalspace = Slab_CONTEXT_HDRSZ(slab->chunksPerBlock); |
| 953 |
|
| 954 |
/* Add the space consumed by blocks in the emptyblocks list */ |
| 955 |
totalspace += dclist_count(&slab->emptyblocks) * slab->blockSize; |
| 956 |
|
| 957 |
for (i = 0; i < SLAB_BLOCKLIST_COUNT; i++) |
| 958 |
{ |
| 959 |
dlist_iter iter; |
| 960 |
|
| 961 |
dlist_foreach(iter, &slab->blocklist[i]) |
| 962 |
{ |
| 963 |
SlabBlock *block = dlist_container(SlabBlock, node, iter.cur); |
| 964 |
|
| 965 |
nblocks++; |
| 966 |
totalspace += slab->blockSize; |
| 967 |
freespace += slab->fullChunkSize * block->nfree; |
| 968 |
freechunks += block->nfree; |
| 969 |
} |
| 970 |
} |
| 971 |
|
| 972 |
if (printfunc) |
| 973 |
{ |
| 974 |
char stats_string[200]; |
| 975 |
|
| 976 |
/* XXX should we include free chunks on empty blocks? */ |
| 977 |
snprintf(stats_string, sizeof(stats_string), |
| 978 |
"%zu total in %zu blocks; %u empty blocks; %zu free (%zu chunks); %zu used", |
| 979 |
totalspace, nblocks, dclist_count(&slab->emptyblocks), |
| 980 |
freespace, freechunks, totalspace - freespace); |
| 981 |
printfunc(context, passthru, stats_string, print_to_stderr); |
| 982 |
} |
| 983 |
|
| 984 |
if (totals) |
| 985 |
{ |
| 986 |
totals->nblocks += nblocks; |
| 987 |
totals->freechunks += freechunks; |
| 988 |
totals->totalspace += totalspace; |
| 989 |
totals->freespace += freespace; |
| 990 |
} |
| 991 |
} |
| 992 |
|
| 993 |
|
| 994 |
#ifdef MEMORY_CONTEXT_CHECKING |
| 995 |
|
| 996 |
/* |
| 997 |
* SlabCheck |
| 998 |
* Walk through all blocks looking for inconsistencies. |
| 999 |
* |
| 1000 |
* NOTE: report errors as WARNING, *not* ERROR or FATAL. Otherwise you'll |
| 1001 |
* find yourself in an infinite loop when trouble occurs, because this |
| 1002 |
* routine will be entered again when elog cleanup tries to release memory! |
| 1003 |
*/ |
| 1004 |
void |
| 1005 |
SlabCheck(MemoryContext context) |
| 1006 |
{ |
| 1007 |
SlabContext *slab = (SlabContext *) context; |
| 1008 |
int i; |
| 1009 |
int nblocks = 0; |
| 1010 |
const char *name = slab->header.name; |
| 1011 |
dlist_iter iter; |
| 1012 |
|
| 1013 |
Assert(SlabIsValid(slab)); |
| 1014 |
Assert(slab->chunksPerBlock > 0); |
| 1015 |
|
| 1016 |
/* |
| 1017 |
* Have a look at the empty blocks. These should have all their chunks |
| 1018 |
* marked as free. Ensure that's the case. |
| 1019 |
*/ |
| 1020 |
dclist_foreach(iter, &slab->emptyblocks) |
| 1021 |
{ |
| 1022 |
SlabBlock *block = dlist_container(SlabBlock, node, iter.cur); |
| 1023 |
|
| 1024 |
if (block->nfree != slab->chunksPerBlock) |
| 1025 |
elog(WARNING, "problem in slab %s: empty block %p should have %d free chunks but has %d chunks free", |
| 1026 |
name, block, slab->chunksPerBlock, block->nfree); |
| 1027 |
} |
| 1028 |
|
| 1029 |
/* walk the non-empty block lists */ |
| 1030 |
for (i = 0; i < SLAB_BLOCKLIST_COUNT; i++) |
| 1031 |
{ |
| 1032 |
int j, |
| 1033 |
nfree; |
| 1034 |
|
| 1035 |
/* walk all blocks on this blocklist */ |
| 1036 |
dlist_foreach(iter, &slab->blocklist[i]) |
| 1037 |
{ |
| 1038 |
SlabBlock *block = dlist_container(SlabBlock, node, iter.cur); |
| 1039 |
MemoryChunk *cur_chunk; |
| 1040 |
|
| 1041 |
/* |
| 1042 |
* Make sure the number of free chunks (in the block header) |
| 1043 |
* matches the position in the blocklist. |
| 1044 |
*/ |
| 1045 |
if (SlabBlocklistIndex(slab, block->nfree) != i) |
| 1046 |
elog(WARNING, "problem in slab %s: block %p is on blocklist %d but should be on blocklist %d", |
| 1047 |
name, block, i, SlabBlocklistIndex(slab, block->nfree)); |
| 1048 |
|
| 1049 |
/* make sure the block is not empty */ |
| 1050 |
if (block->nfree >= slab->chunksPerBlock) |
| 1051 |
elog(WARNING, "problem in slab %s: empty block %p incorrectly stored on blocklist element %d", |
| 1052 |
name, block, i); |
| 1053 |
|
| 1054 |
/* make sure the slab pointer correctly points to this context */ |
| 1055 |
if (block->slab != slab) |
| 1056 |
elog(WARNING, "problem in slab %s: bogus slab link in block %p", |
| 1057 |
name, block); |
| 1058 |
|
| 1059 |
/* reset the array of free chunks for this block */ |
| 1060 |
memset(slab->isChunkFree, 0, (slab->chunksPerBlock * sizeof(bool))); |
| 1061 |
nfree = 0; |
| 1062 |
|
| 1063 |
/* walk through the block's free list chunks */ |
| 1064 |
cur_chunk = block->freehead; |
| 1065 |
while (cur_chunk != NULL) |
| 1066 |
{ |
| 1067 |
int chunkidx = SlabChunkIndex(slab, block, cur_chunk); |
| 1068 |
|
| 1069 |
/* |
| 1070 |
* Ensure the free list link points to something on the block |
| 1071 |
* at an address aligned according to the full chunk size. |
| 1072 |
*/ |
| 1073 |
if (cur_chunk < SlabBlockGetChunk(slab, block, 0) || |
| 1074 |
cur_chunk > SlabBlockGetChunk(slab, block, slab->chunksPerBlock - 1) || |
| 1075 |
SlabChunkMod(slab, block, cur_chunk) != 0) |
| 1076 |
elog(WARNING, "problem in slab %s: bogus free list link %p in block %p", |
| 1077 |
name, cur_chunk, block); |
| 1078 |
|
| 1079 |
/* count the chunk and mark it free on the free chunk array */ |
| 1080 |
nfree++; |
| 1081 |
slab->isChunkFree[chunkidx] = true; |
| 1082 |
|
| 1083 |
/* read pointer of the next free chunk */ |
| 1084 |
VALGRIND_MAKE_MEM_DEFINED(MemoryChunkGetPointer(cur_chunk), sizeof(MemoryChunk *)); |
| 1085 |
cur_chunk = *(MemoryChunk **) SlabChunkGetPointer(cur_chunk); |
| 1086 |
} |
| 1087 |
|
| 1088 |
/* check that the unused pointer matches what nunused claims */ |
| 1089 |
if (SlabBlockGetChunk(slab, block, slab->chunksPerBlock - block->nunused) != |
| 1090 |
block->unused) |
| 1091 |
elog(WARNING, "problem in slab %s: mismatch detected between nunused chunks and unused pointer in block %p", |
| 1092 |
name, block); |
| 1093 |
|
| 1094 |
/* |
| 1095 |
* count the remaining free chunks that have yet to make it onto |
| 1096 |
* the block's free list. |
| 1097 |
*/ |
| 1098 |
cur_chunk = block->unused; |
| 1099 |
for (j = 0; j < block->nunused; j++) |
| 1100 |
{ |
| 1101 |
int chunkidx = SlabChunkIndex(slab, block, cur_chunk); |
| 1102 |
|
| 1103 |
|
| 1104 |
/* count the chunk as free and mark it as so in the array */ |
| 1105 |
nfree++; |
| 1106 |
if (chunkidx < slab->chunksPerBlock) |
| 1107 |
slab->isChunkFree[chunkidx] = true; |
| 1108 |
|
| 1109 |
/* move forward 1 chunk */ |
| 1110 |
cur_chunk = (MemoryChunk *) (((char *) cur_chunk) + slab->fullChunkSize); |
| 1111 |
} |
| 1112 |
|
| 1113 |
for (j = 0; j < slab->chunksPerBlock; j++) |
| 1114 |
{ |
| 1115 |
if (!slab->isChunkFree[j]) |
| 1116 |
{ |
| 1117 |
MemoryChunk *chunk = SlabBlockGetChunk(slab, block, j); |
| 1118 |
SlabBlock *chunkblock; |
| 1119 |
|
| 1120 |
/* Allow access to the chunk header. */ |
| 1121 |
VALGRIND_MAKE_MEM_DEFINED(chunk, Slab_CHUNKHDRSZ); |
| 1122 |
|
| 1123 |
chunkblock = (SlabBlock *) MemoryChunkGetBlock(chunk); |
| 1124 |
|
| 1125 |
/* Disallow access to the chunk header. */ |
| 1126 |
VALGRIND_MAKE_MEM_NOACCESS(chunk, Slab_CHUNKHDRSZ); |
| 1127 |
|
| 1128 |
/* |
| 1129 |
* check the chunk's blockoffset correctly points back to |
| 1130 |
* the block |
| 1131 |
*/ |
| 1132 |
if (chunkblock != block) |
| 1133 |
elog(WARNING, "problem in slab %s: bogus block link in block %p, chunk %p", |
| 1134 |
name, block, chunk); |
| 1135 |
|
| 1136 |
/* check the sentinel byte is intact */ |
| 1137 |
Assert(slab->chunkSize < (slab->fullChunkSize - Slab_CHUNKHDRSZ)); |
| 1138 |
if (!sentinel_ok(chunk, Slab_CHUNKHDRSZ + slab->chunkSize)) |
| 1139 |
elog(WARNING, "problem in slab %s: detected write past chunk end in block %p, chunk %p", |
| 1140 |
name, block, chunk); |
| 1141 |
} |
| 1142 |
} |
| 1143 |
|
| 1144 |
/* |
| 1145 |
* Make sure we got the expected number of free chunks (as tracked |
| 1146 |
* in the block header). |
| 1147 |
*/ |
| 1148 |
if (nfree != block->nfree) |
| 1149 |
elog(WARNING, "problem in slab %s: nfree in block %p is %d but %d chunk were found as free", |
| 1150 |
name, block, block->nfree, nfree); |
| 1151 |
|
| 1152 |
nblocks++; |
| 1153 |
} |
| 1154 |
} |
| 1155 |
|
| 1156 |
/* the stored empty blocks are tracked in mem_allocated too */ |
| 1157 |
nblocks += dclist_count(&slab->emptyblocks); |
| 1158 |
|
| 1159 |
Assert(nblocks * slab->blockSize == context->mem_allocated); |
| 1160 |
} |
| 1161 |
|
| 1162 |
#endif /* MEMORY_CONTEXT_CHECKING */ |