shmem.c

Go to the documentation of this file.

/*-------------------------------------------------------------------------
 *
 * shmem.c
 *    create shared memory and initialize shared memory data structures.
 *
 * Portions Copyright (c) 1996-2026, PostgreSQL Global Development Group
 * Portions Copyright (c) 1994, Regents of the University of California
 *
 *
 * IDENTIFICATION
 *    src/backend/storage/ipc/shmem.c
 *
 *-------------------------------------------------------------------------
 */
/*
 * POSTGRES processes share one or more regions of shared memory.
 * The shared memory is created by a postmaster and is inherited
 * by each backend via fork() (or, in some ports, via other OS-specific
 * methods).  The routines in this file are used for allocating and
 * binding to shared memory data structures.
 *
 * NOTES:
 *      (a) There are three kinds of shared memory data structures
 *  available to POSTGRES: fixed-size structures, queues and hash
 *  tables.  Fixed-size structures contain things like global variables
 *  for a module and should never be allocated after the shared memory
 *  initialization phase.  Hash tables have a fixed maximum size, but
 *  their actual size can vary dynamically.  When entries are added
 *  to the table, more space is allocated.  Queues link data structures
 *  that have been allocated either within fixed-size structures or as hash
 *  buckets.  Each shared data structure has a string name to identify
 *  it (assigned in the module that declares it).
 *
 *      (b) During initialization, each module looks for its
 *  shared data structures in a hash table called the "Shmem Index".
 *  If the data structure is not present, the caller can allocate
 *  a new one and initialize it.  If the data structure is present,
 *  the caller "attaches" to the structure by initializing a pointer
 *  in the local address space.
 *      The shmem index has two purposes: first, it gives us
 *  a simple model of how the world looks when a backend process
 *  initializes.  If something is present in the shmem index,
 *  it is initialized.  If it is not, it is uninitialized.  Second,
 *  the shmem index allows us to allocate shared memory on demand
 *  instead of trying to preallocate structures and hard-wire the
 *  sizes and locations in header files.  If you are using a lot
 *  of shared memory in a lot of different places (and changing
 *  things during development), this is important.
 *
 *      (c) In standard Unix-ish environments, individual backends do not
 *  need to re-establish their local pointers into shared memory, because
 *  they inherit correct values of those variables via fork() from the
 *  postmaster.  However, this does not work in the EXEC_BACKEND case.
 *  In ports using EXEC_BACKEND, new backends have to set up their local
 *  pointers using the method described in (b) above.
 *
 *      (d) memory allocation model: shared memory can never be
 *  freed, once allocated.   Each hash table has its own free list,
 *  so hash buckets can be reused when an item is deleted.  However,
 *  if one hash table grows very large and then shrinks, its space
 *  cannot be redistributed to other tables.  We could build a simple
 *  hash bucket garbage collector if need be.  Right now, it seems
 *  unnecessary.
 */
 
#include "postgres.h"
 
#include <unistd.h>
 
#include "common/int.h"
#include "fmgr.h"
#include "funcapi.h"
#include "miscadmin.h"
#include "port/pg_numa.h"
#include "storage/lwlock.h"
#include "storage/pg_shmem.h"
#include "storage/shmem.h"
#include "storage/spin.h"
#include "utils/builtins.h"
#include "utils/tuplestore.h"
 
/*
 * This is the first data structure stored in the shared memory segment, at
 * the offset that PGShmemHeader->content_offset points to.  Allocations by
 * ShmemAlloc() are carved out of the space after this.
 *
 * For the base pointer and the total size of the shmem segment, we rely on
 * the PGShmemHeader.
 */
typedef struct ShmemAllocatorData
{
    Size        free_offset;    /* offset to first free space from ShmemBase */
    HASHHDR    *index;          /* location of ShmemIndex */
 
    /* protects shared memory and LWLock allocation */
    slock_t     shmem_lock;
} ShmemAllocatorData;
 
static void *ShmemAllocRaw(Size size, Size *allocated_size);
 
/* shared memory global variables */
 
static PGShmemHeader *ShmemSegHdr;  /* shared mem segment header */
static void *ShmemBase;         /* start address of shared memory */
static void *ShmemEnd;          /* end+1 address of shared memory */
 
static ShmemAllocatorData *ShmemAllocator;
slock_t    *ShmemLock;          /* points to ShmemAllocator->shmem_lock */
static HTAB *ShmemIndex = NULL; /* primary index hashtable for shmem */
 
/* To get reliable results for NUMA inquiry we need to "touch pages" once */
static bool firstNumaTouch = true;
 
Datum       pg_numa_available(PG_FUNCTION_ARGS);
 
/*
 *  InitShmemAllocator() --- set up basic pointers to shared memory.
 *
 * Called at postmaster or stand-alone backend startup, to initialize the
 * allocator's data structure in the shared memory segment.  In EXEC_BACKEND,
 * this is also called at backend startup, to set up pointers to the shared
 * memory areas.
 */
void
InitShmemAllocator(PGShmemHeader *seghdr)
{
    Assert(seghdr != NULL);
 
    /*
     * We assume the pointer and offset are MAXALIGN.  Not a hard requirement,
     * but it's true today and keeps the math below simpler.
     */
    Assert(seghdr == (void *) MAXALIGN(seghdr));
    Assert(seghdr->content_offset == MAXALIGN(seghdr->content_offset));
 
    ShmemSegHdr = seghdr;
    ShmemBase = seghdr;
    ShmemEnd = (char *) ShmemBase + seghdr->totalsize;
 
#ifndef EXEC_BACKEND
    Assert(!IsUnderPostmaster);
#endif
    if (IsUnderPostmaster)
    {
        PGShmemHeader *shmhdr = ShmemSegHdr;
 
        ShmemAllocator = (ShmemAllocatorData *) ((char *) shmhdr + shmhdr->content_offset);
        ShmemLock = &ShmemAllocator->shmem_lock;
    }
    else
    {
        Size        offset;
 
        /*
         * Allocations after this point should go through ShmemAlloc, which
         * expects to allocate everything on cache line boundaries.  Make sure
         * the first allocation begins on a cache line boundary.
         */
        offset = CACHELINEALIGN(seghdr->content_offset + sizeof(ShmemAllocatorData));
        if (offset > seghdr->totalsize)
            ereport(ERROR,
                    (errcode(ERRCODE_OUT_OF_MEMORY),
                     errmsg("out of shared memory (%zu bytes requested)",
                            offset)));
 
        ShmemAllocator = (ShmemAllocatorData *) ((char *) seghdr + seghdr->content_offset);
 
        SpinLockInit(&ShmemAllocator->shmem_lock);
        ShmemLock = &ShmemAllocator->shmem_lock;
        ShmemAllocator->free_offset = offset;
        /* ShmemIndex can't be set up yet (need LWLocks first) */
        ShmemAllocator->index = NULL;
        ShmemIndex = (HTAB *) NULL;
    }
}
 
/*
 * ShmemAlloc -- allocate max-aligned chunk from shared memory
 *
 * Throws error if request cannot be satisfied.
 *
 * Assumes ShmemLock and ShmemSegHdr are initialized.
 */
void *
ShmemAlloc(Size size)
{
    void       *newSpace;
    Size        allocated_size;
 
    newSpace = ShmemAllocRaw(size, &allocated_size);
    if (!newSpace)
        ereport(ERROR,
                (errcode(ERRCODE_OUT_OF_MEMORY),
                 errmsg("out of shared memory (%zu bytes requested)",
                        size)));
    return newSpace;
}
 
/*
 * ShmemAllocNoError -- allocate max-aligned chunk from shared memory
 *
 * As ShmemAlloc, but returns NULL if out of space, rather than erroring.
 */
void *
ShmemAllocNoError(Size size)
{
    Size        allocated_size;
 
    return ShmemAllocRaw(size, &allocated_size);
}
 
/*
 * ShmemAllocRaw -- allocate align chunk and return allocated size
 *
 * Also sets *allocated_size to the number of bytes allocated, which will
 * be equal to the number requested plus any padding we choose to add.
 */
static void *
ShmemAllocRaw(Size size, Size *allocated_size)
{
    Size        newStart;
    Size        newFree;
    void       *newSpace;
 
    /*
     * Ensure all space is adequately aligned.  We used to only MAXALIGN this
     * space but experience has proved that on modern systems that is not good
     * enough.  Many parts of the system are very sensitive to critical data
     * structures getting split across cache line boundaries.  To avoid that,
     * attempt to align the beginning of the allocation to a cache line
     * boundary.  The calling code will still need to be careful about how it
     * uses the allocated space - e.g. by padding each element in an array of
     * structures out to a power-of-two size - but without this, even that
     * won't be sufficient.
     */
    size = CACHELINEALIGN(size);
    *allocated_size = size;
 
    Assert(ShmemSegHdr != NULL);
 
    SpinLockAcquire(ShmemLock);
 
    newStart = ShmemAllocator->free_offset;
 
    newFree = newStart + size;
    if (newFree <= ShmemSegHdr->totalsize)
    {
        newSpace = (char *) ShmemBase + newStart;
        ShmemAllocator->free_offset = newFree;
    }
    else
        newSpace = NULL;
 
    SpinLockRelease(ShmemLock);
 
    /* note this assert is okay with newSpace == NULL */
    Assert(newSpace == (void *) CACHELINEALIGN(newSpace));
 
    return newSpace;
}
 
/*
 * ShmemAddrIsValid -- test if an address refers to shared memory
 *
 * Returns true if the pointer points within the shared memory segment.
 */
bool
ShmemAddrIsValid(const void *addr)
{
    return (addr >= ShmemBase) && (addr < ShmemEnd);
}
 
/*
 *  InitShmemIndex() --- set up or attach to shmem index table.
 */
void
InitShmemIndex(void)
{
    HASHCTL     info;
 
    /*
     * Create the shared memory shmem index.
     *
     * Since ShmemInitHash calls ShmemInitStruct, which expects the ShmemIndex
     * hashtable to exist already, we have a bit of a circularity problem in
     * initializing the ShmemIndex itself.  The special "ShmemIndex" hash
     * table name will tell ShmemInitStruct to fake it.
     */
    info.keysize = SHMEM_INDEX_KEYSIZE;
    info.entrysize = sizeof(ShmemIndexEnt);
 
    ShmemIndex = ShmemInitHash("ShmemIndex",
                               SHMEM_INDEX_SIZE, SHMEM_INDEX_SIZE,
                               &info,
                               HASH_ELEM | HASH_STRINGS);
}
 
/*
 * ShmemInitHash -- Create and initialize, or attach to, a
 *      shared memory hash table.
 *
 * We assume caller is doing some kind of synchronization
 * so that two processes don't try to create/initialize the same
 * table at once.  (In practice, all creations are done in the postmaster
 * process; child processes should always be attaching to existing tables.)
 *
 * max_size is the estimated maximum number of hashtable entries.  This is
 * not a hard limit, but the access efficiency will degrade if it is
 * exceeded substantially (since it's used to compute directory size and
 * the hash table buckets will get overfull).
 *
 * init_size is the number of hashtable entries to preallocate.  For a table
 * whose maximum size is certain, this should be equal to max_size; that
 * ensures that no run-time out-of-shared-memory failures can occur.
 *
 * *infoP and hash_flags must specify at least the entry sizes and key
 * comparison semantics (see hash_create()).  Flag bits and values specific
 * to shared-memory hash tables are added here, except that callers may
 * choose to specify HASH_PARTITION and/or HASH_FIXED_SIZE.
 *
 * Note: before Postgres 9.0, this function returned NULL for some failure
 * cases.  Now, it always throws error instead, so callers need not check
 * for NULL.
 */
HTAB *
ShmemInitHash(const char *name,     /* table string name for shmem index */
              int64 init_size,  /* initial table size */
              int64 max_size,   /* max size of the table */
              HASHCTL *infoP,   /* info about key and bucket size */
              int hash_flags)   /* info about infoP */
{
    bool        found;
    void       *location;
 
    /*
     * Hash tables allocated in shared memory have a fixed directory; it can't
     * grow or other backends wouldn't be able to find it. So, make sure we
     * make it big enough to start with.
     *
     * The shared memory allocator must be specified too.
     */
    infoP->dsize = infoP->max_dsize = hash_select_dirsize(max_size);
    infoP->alloc = ShmemAllocNoError;
    hash_flags |= HASH_SHARED_MEM | HASH_ALLOC | HASH_DIRSIZE;
 
    /* look it up in the shmem index */
    location = ShmemInitStruct(name,
                               hash_get_shared_size(infoP, hash_flags),
                               &found);
 
    /*
     * if it already exists, attach to it rather than allocate and initialize
     * new space
     */
    if (found)
        hash_flags |= HASH_ATTACH;
 
    /* Pass location of hashtable header to hash_create */
    infoP->hctl = (HASHHDR *) location;
 
    return hash_create(name, init_size, infoP, hash_flags);
}
 
/*
 * ShmemInitStruct -- Create/attach to a structure in shared memory.
 *
 *      This is called during initialization to find or allocate
 *      a data structure in shared memory.  If no other process
 *      has created the structure, this routine allocates space
 *      for it.  If it exists already, a pointer to the existing
 *      structure is returned.
 *
 *  Returns: pointer to the object.  *foundPtr is set true if the object was
 *      already in the shmem index (hence, already initialized).
 *
 *  Note: before Postgres 9.0, this function returned NULL for some failure
 *  cases.  Now, it always throws error instead, so callers need not check
 *  for NULL.
 */
void *
ShmemInitStruct(const char *name, Size size, bool *foundPtr)
{
    ShmemIndexEnt *result;
    void       *structPtr;
 
    LWLockAcquire(ShmemIndexLock, LW_EXCLUSIVE);
 
    if (!ShmemIndex)
    {
        /* Must be trying to create/attach to ShmemIndex itself */
        Assert(strcmp(name, "ShmemIndex") == 0);
 
        if (IsUnderPostmaster)
        {
            /* Must be initializing a (non-standalone) backend */
            Assert(ShmemAllocator->index != NULL);
            structPtr = ShmemAllocator->index;
            *foundPtr = true;
        }
        else
        {
            /*
             * If the shmem index doesn't exist, we are bootstrapping: we must
             * be trying to init the shmem index itself.
             *
             * Notice that the ShmemIndexLock is released before the shmem
             * index has been initialized.  This should be OK because no other
             * process can be accessing shared memory yet.
             */
            Assert(ShmemAllocator->index == NULL);
            structPtr = ShmemAlloc(size);
            ShmemAllocator->index = structPtr;
            *foundPtr = false;
        }
        LWLockRelease(ShmemIndexLock);
        return structPtr;
    }
 
    /* look it up in the shmem index */
    result = (ShmemIndexEnt *)
        hash_search(ShmemIndex, name, HASH_ENTER_NULL, foundPtr);
 
    if (!result)
    {
        LWLockRelease(ShmemIndexLock);
        ereport(ERROR,
                (errcode(ERRCODE_OUT_OF_MEMORY),
                 errmsg("could not create ShmemIndex entry for data structure \"%s\"",
                        name)));
    }
 
    if (*foundPtr)
    {
        /*
         * Structure is in the shmem index so someone else has allocated it
         * already.  The size better be the same as the size we are trying to
         * initialize to, or there is a name conflict (or worse).
         */
        if (result->size != size)
        {
            LWLockRelease(ShmemIndexLock);
            ereport(ERROR,
                    (errmsg("ShmemIndex entry size is wrong for data structure"
                            " \"%s\": expected %zu, actual %zu",
                            name, size, result->size)));
        }
        structPtr = result->location;
    }
    else
    {
        Size        allocated_size;
 
        /* It isn't in the table yet. allocate and initialize it */
        structPtr = ShmemAllocRaw(size, &allocated_size);
        if (structPtr == NULL)
        {
            /* out of memory; remove the failed ShmemIndex entry */
            hash_search(ShmemIndex, name, HASH_REMOVE, NULL);
            LWLockRelease(ShmemIndexLock);
            ereport(ERROR,
                    (errcode(ERRCODE_OUT_OF_MEMORY),
                     errmsg("not enough shared memory for data structure"
                            " \"%s\" (%zu bytes requested)",
                            name, size)));
        }
        result->size = size;
        result->allocated_size = allocated_size;
        result->location = structPtr;
    }
 
    LWLockRelease(ShmemIndexLock);
 
    Assert(ShmemAddrIsValid(structPtr));
 
    Assert(structPtr == (void *) CACHELINEALIGN(structPtr));
 
    return structPtr;
}
 
 
/*
 * Add two Size values, checking for overflow
 */
Size
add_size(Size s1, Size s2)
{
    Size        result;
 
    if (pg_add_size_overflow(s1, s2, &result))
        ereport(ERROR,
                (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
                 errmsg("requested shared memory size overflows size_t")));
    return result;
}
 
/*
 * Multiply two Size values, checking for overflow
 */
Size
mul_size(Size s1, Size s2)
{
    Size        result;
 
    if (pg_mul_size_overflow(s1, s2, &result))
        ereport(ERROR,
                (errcode(ERRCODE_PROGRAM_LIMIT_EXCEEDED),
                 errmsg("requested shared memory size overflows size_t")));
    return result;
}
 
/* SQL SRF showing allocated shared memory */
Datum
pg_get_shmem_allocations(PG_FUNCTION_ARGS)
{
#define PG_GET_SHMEM_SIZES_COLS 4
    ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
    HASH_SEQ_STATUS hstat;
    ShmemIndexEnt *ent;
    Size        named_allocated = 0;
    Datum       values[PG_GET_SHMEM_SIZES_COLS];
    bool        nulls[PG_GET_SHMEM_SIZES_COLS];
 
    InitMaterializedSRF(fcinfo, 0);
 
    LWLockAcquire(ShmemIndexLock, LW_SHARED);
 
    hash_seq_init(&hstat, ShmemIndex);
 
    /* output all allocated entries */
    memset(nulls, 0, sizeof(nulls));
    while ((ent = (ShmemIndexEnt *) hash_seq_search(&hstat)) != NULL)
    {
        values[0] = CStringGetTextDatum(ent->key);
        values[1] = Int64GetDatum((char *) ent->location - (char *) ShmemSegHdr);
        values[2] = Int64GetDatum(ent->size);
        values[3] = Int64GetDatum(ent->allocated_size);
        named_allocated += ent->allocated_size;
 
        tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
                             values, nulls);
    }
 
    /* output shared memory allocated but not counted via the shmem index */
    values[0] = CStringGetTextDatum("<anonymous>");
    nulls[1] = true;
    values[2] = Int64GetDatum(ShmemAllocator->free_offset - named_allocated);
    values[3] = values[2];
    tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
 
    /* output as-of-yet unused shared memory */
    nulls[0] = true;
    values[1] = Int64GetDatum(ShmemAllocator->free_offset);
    nulls[1] = false;
    values[2] = Int64GetDatum(ShmemSegHdr->totalsize - ShmemAllocator->free_offset);
    values[3] = values[2];
    tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc, values, nulls);
 
    LWLockRelease(ShmemIndexLock);
 
    return (Datum) 0;
}
 
/*
 * SQL SRF showing NUMA memory nodes for allocated shared memory
 *
 * Compared to pg_get_shmem_allocations(), this function does not return
 * information about shared anonymous allocations and unused shared memory.
 */
Datum
pg_get_shmem_allocations_numa(PG_FUNCTION_ARGS)
{
#define PG_GET_SHMEM_NUMA_SIZES_COLS 3
    ReturnSetInfo *rsinfo = (ReturnSetInfo *) fcinfo->resultinfo;
    HASH_SEQ_STATUS hstat;
    ShmemIndexEnt *ent;
    Datum       values[PG_GET_SHMEM_NUMA_SIZES_COLS];
    bool        nulls[PG_GET_SHMEM_NUMA_SIZES_COLS];
    Size        os_page_size;
    void      **page_ptrs;
    int        *pages_status;
    uint64      shm_total_page_count,
                shm_ent_page_count,
                max_nodes;
    Size       *nodes;
 
    if (pg_numa_init() == -1)
        elog(ERROR, "libnuma initialization failed or NUMA is not supported on this platform");
 
    InitMaterializedSRF(fcinfo, 0);
 
    max_nodes = pg_numa_get_max_node();
    nodes = palloc_array(Size, max_nodes + 2);
 
    /*
     * Shared memory allocations can vary in size and may not align with OS
     * memory page boundaries, while NUMA queries work on pages.
     *
     * To correctly map each allocation to NUMA nodes, we need to: 1.
     * Determine the OS memory page size. 2. Align each allocation's start/end
     * addresses to page boundaries. 3. Query NUMA node information for all
     * pages spanning the allocation.
     */
    os_page_size = pg_get_shmem_pagesize();
 
    /*
     * Allocate memory for page pointers and status based on total shared
     * memory size. This simplified approach allocates enough space for all
     * pages in shared memory rather than calculating the exact requirements
     * for each segment.
     *
     * Add 1, because we don't know how exactly the segments align to OS
     * pages, so the allocation might use one more memory page. In practice
     * this is not very likely, and moreover we have more entries, each of
     * them using only fraction of the total pages.
     */
    shm_total_page_count = (ShmemSegHdr->totalsize / os_page_size) + 1;
    page_ptrs = palloc0_array(void *, shm_total_page_count);
    pages_status = palloc_array(int, shm_total_page_count);
 
    if (firstNumaTouch)
        elog(DEBUG1, "NUMA: page-faulting shared memory segments for proper NUMA readouts");
 
    LWLockAcquire(ShmemIndexLock, LW_SHARED);
 
    hash_seq_init(&hstat, ShmemIndex);
 
    /* output all allocated entries */
    while ((ent = (ShmemIndexEnt *) hash_seq_search(&hstat)) != NULL)
    {
        int         i;
        char       *startptr,
                   *endptr;
        Size        total_len;
 
        /*
         * Calculate the range of OS pages used by this segment. The segment
         * may start / end half-way through a page, we want to count these
         * pages too. So we align the start/end pointers down/up, and then
         * calculate the number of pages from that.
         */
        startptr = (char *) TYPEALIGN_DOWN(os_page_size, ent->location);
        endptr = (char *) TYPEALIGN(os_page_size,
                                    (char *) ent->location + ent->allocated_size);
        total_len = (endptr - startptr);
 
        shm_ent_page_count = total_len / os_page_size;
 
        /*
         * If we ever get 0xff (-1) back from kernel inquiry, then we probably
         * have a bug in mapping buffers to OS pages.
         */
        memset(pages_status, 0xff, sizeof(int) * shm_ent_page_count);
 
        /*
         * Setup page_ptrs[] with pointers to all OS pages for this segment,
         * and get the NUMA status using pg_numa_query_pages.
         *
         * In order to get reliable results we also need to touch memory
         * pages, so that inquiry about NUMA memory node doesn't return -2
         * (ENOENT, which indicates unmapped/unallocated pages).
         */
        for (i = 0; i < shm_ent_page_count; i++)
        {
            page_ptrs[i] = startptr + (i * os_page_size);
 
            if (firstNumaTouch)
                pg_numa_touch_mem_if_required(page_ptrs[i]);
 
            CHECK_FOR_INTERRUPTS();
        }
 
        if (pg_numa_query_pages(0, shm_ent_page_count, page_ptrs, pages_status) == -1)
            elog(ERROR, "failed NUMA pages inquiry status: %m");
 
        /* Count number of NUMA nodes used for this shared memory entry */
        memset(nodes, 0, sizeof(Size) * (max_nodes + 2));
 
        for (i = 0; i < shm_ent_page_count; i++)
        {
            int         s = pages_status[i];
 
            /* Ensure we are adding only valid index to the array */
            if (s >= 0 && s <= max_nodes)
            {
                /* valid NUMA node */
                nodes[s]++;
                continue;
            }
            else if (s == -2)
            {
                /* -2 means ENOENT (e.g. page was moved to swap) */
                nodes[max_nodes + 1]++;
                continue;
            }
 
            elog(ERROR, "invalid NUMA node id outside of allowed range "
                 "[0, " UINT64_FORMAT "]: %d", max_nodes, s);
        }
 
        /* no NULLs for regular nodes */
        memset(nulls, 0, sizeof(nulls));
 
        /*
         * Add one entry for each NUMA node, including those without allocated
         * memory for this segment.
         */
        for (i = 0; i <= max_nodes; i++)
        {
            values[0] = CStringGetTextDatum(ent->key);
            values[1] = Int32GetDatum(i);
            values[2] = Int64GetDatum(nodes[i] * os_page_size);
 
            tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
                                 values, nulls);
        }
 
        /* The last entry is used for pages without a NUMA node. */
        nulls[1] = true;
        values[0] = CStringGetTextDatum(ent->key);
        values[2] = Int64GetDatum(nodes[max_nodes + 1] * os_page_size);
 
        tuplestore_putvalues(rsinfo->setResult, rsinfo->setDesc,
                             values, nulls);
    }
 
    LWLockRelease(ShmemIndexLock);
    firstNumaTouch = false;
 
    return (Datum) 0;
}
 
/*
 * Determine the memory page size used for the shared memory segment.
 *
 * If the shared segment was allocated using huge pages, returns the size of
 * a huge page. Otherwise returns the size of regular memory page.
 *
 * This should be used only after the server is started.
 */
Size
pg_get_shmem_pagesize(void)
{
    Size        os_page_size;
#ifdef WIN32
    SYSTEM_INFO sysinfo;
 
    GetSystemInfo(&sysinfo);
    os_page_size = sysinfo.dwPageSize;
#else
    os_page_size = sysconf(_SC_PAGESIZE);
#endif
 
    Assert(IsUnderPostmaster);
    Assert(huge_pages_status != HUGE_PAGES_UNKNOWN);
 
    if (huge_pages_status == HUGE_PAGES_ON)
        GetHugePageSize(&os_page_size, NULL);
 
    return os_page_size;
}
 
Datum
pg_numa_available(PG_FUNCTION_ARGS)
{
    PG_RETURN_BOOL(pg_numa_init() != -1);
}