Re: Re: [HACKERS] Patch für MAP_HUGETLB for mmap() shared memory

Did we decide against specifying huge pages in Postgres?

---------------------------------------------------------------------------

On Tue, Oct 30, 2012 at 09:16:07PM +0100, Christian Kruse wrote:
> Hey,
>
> ok, I think I implemented all of the changes you requested. All but
> the ia64 dependent, I have to do more research for this one.
>
>
> Greetings,
> CK

> diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
> index b4fcbaf..66ed10f 100644
> --- a/doc/src/sgml/config.sgml
> +++ b/doc/src/sgml/config.sgml
> @@ -1049,6 +1049,37 @@ include 'filename'
> </listitem>
> </varlistentry>
>
> + <varlistentry id="guc-huge-tlb-pages" xreflabel="huge_tlb_pages">
> + <term><varname>huge_tlb_pages</varname> (<type>enum</type>)</term>
> + <indexterm>
> + <primary><varname>huge_tlb_pages</> configuration parameter</primary>
> + </indexterm>
> + <listitem>
> + <para>
> + Enables/disables the use of huge tlb pages. Valid values are
> + <literal>on</literal>, <literal>off</literal> and <literal>try</literal>.
> + The default value is <literal>try</literal>.
> + </para>
> +
> + <para>
> + With <varname>huge_tlb_pages</varname> set to <literal>on</literal>
> + <symbol>mmap()</symbol> will be called with <symbol>MAP_HUGETLB</symbol>.
> + If the call fails the server will fail fatally.
> + </para>
> +
> + <para>
> + With <varname>huge_tlb_pages</varname> set to <literal>off</literal> we
> + will not use <symbol>MAP_HUGETLB</symbol> at all.
> + </para>
> +
> + <para>
> + With <varname>huge_tlb_pages</varname> set to <literal>try</literal>
> + we will try to use <symbol>MAP_HUGETLB</symbol> and fall back to
> + <symbol>mmap()</symbol> without <symbol>MAP_HUGETLB</symbol>.
> + </para>
> + </listitem>
> + </varlistentry>
> +
> <varlistentry id="guc-temp-buffers" xreflabel="temp_buffers">
> <term><varname>temp_buffers</varname> (<type>integer</type>)</term>
> <indexterm>
> diff --git a/src/backend/port/sysv_shmem.c b/src/backend/port/sysv_shmem.c
> index df06312..f9de239 100644
> --- a/src/backend/port/sysv_shmem.c
> +++ b/src/backend/port/sysv_shmem.c
> @@ -27,10 +27,14 @@
> #ifdef HAVE_SYS_SHM_H
> #include <sys/shm.h>
> #endif
> +#ifdef MAP_HUGETLB
> +#include <dirent.h>
> +#endif
>
> #include "miscadmin.h"
> #include "storage/ipc.h"
> #include "storage/pg_shmem.h"
> +#include "utils/guc.h"
>
>
> typedef key_t IpcMemoryKey; /* shared memory key passed to shmget(2) */
> @@ -61,6 +65,19 @@ typedef int IpcMemoryId; /* shared memory ID returned by shmget(2) */
> #define MAP_FAILED ((void *) -1)
> #endif
>
> +#ifdef MAP_HUGETLB
> +# ifdef __ia64__
> +# define PG_HUGETLB_BASE_ADDR (void *)(0x8000000000000000UL)
> +# define PG_MAP_HUGETLB (MAP_HUGETLB|MAP_FIXED)
> +# else
> +# define PG_HUGETLB_BASE_ADDR (void *)(0x0UL)
> +# define PG_MAP_HUGETLB MAP_HUGETLB
> +# endif
> +#else
> +# define PG_MAP_HUGETLB 0
> +#endif
> +
> +
>
> unsigned long UsedShmemSegID = 0;
> void *UsedShmemSegAddr = NULL;
> @@ -73,7 +90,6 @@ static void IpcMemoryDelete(int status, Datum shmId);
> static PGShmemHeader *PGSharedMemoryAttach(IpcMemoryKey key,
> IpcMemoryId *shmid);
>
> -
> /*
> * InternalIpcMemoryCreate(memKey, size)
> *
> @@ -342,6 +358,155 @@ PGSharedMemoryIsInUse(unsigned long id1, unsigned long id2)
> }
>
>
> +#ifdef MAP_HUGETLB
> +#define HUGE_PAGE_INFO_DIR "/sys/kernel/mm/hugepages"
> +
> +/*
> + * static long InternalGetFreeHugepagesCount(const char *name)
> + *
> + * Attempt to read the number of available hugepages from
> + * /sys/kernel/mm/hugepages/hugepages-<size>/free_hugepages
> + * Will fail (return -1) if file could not be opened, 0 if no pages are available
> + * and > 0 if there are free pages
> + *
> + */
> +static long
> +InternalGetFreeHugepagesCount(const char *name)
> +{
> + int fd;
> + char buff[1024];
> + size_t len;
> + long result;
> + char *ptr;
> +
> + len = snprintf(buff, 1024, "%s/%s/free_hugepages", HUGE_PAGE_INFO_DIR, name);
> + if (len == 1024) /* I don't think that this will happen ever */
> + {
> + ereport(huge_tlb_pages == HUGE_TLB_TRY ? DEBUG1 : WARNING,
> + (errmsg("Filename %s/%s/free_hugepages is too long", HUGE_PAGE_INFO_DIR, name),
> + errcontext("while checking hugepage size")));
> + return -1;
> + }
> +
> + fd = open(buff, O_RDONLY);
> + if (fd <= 0)
> + {
> + ereport(huge_tlb_pages == HUGE_TLB_TRY ? DEBUG1 : WARNING,
> + (errmsg("Could not open file %s: %s", buff, strerror(errno)),
> + errcontext("while checking hugepage size")));
> + return -1;
> + }
> +
> + len = read(fd, buff, 1024);
> + if (len <= 0)
> + {
> + ereport(huge_tlb_pages == HUGE_TLB_TRY ? DEBUG1 : WARNING,
> + (errmsg("Error reading from file %s: %s", buff, strerror(errno)),
> + errcontext("while checking hugepage size")));
> + close(fd);
> + return -1;
> + }
> +
> + /*
> + * If the content of free_hugepages is longer than or equal to 1024 bytes
> + * the rest is irrelevant; we simply want to know if there are any
> + * hugepages left
> + */
> + if (len == 1024)
> + {
> + buff[1023] = 0;
> + }
> + else
> + {
> + buff[len] = 0;
> + }
> +
> + close(fd);
> +
> + result = strtol(buff, &ptr, 10);
> +
> + if (ptr == NULL)
> + {
> + ereport(huge_tlb_pages == HUGE_TLB_TRY ? DEBUG1 : WARNING,
> + (errmsg("Could not convert contents of file %s/%s/free_hugepages to number", HUGE_PAGE_INFO_DIR, name),
> + errcontext("while checking hugepage size")));
> + return -1;
> + }
> +
> + return result;
> +}
> +
> +/*
> + * static long InternalGetHugepageSize()
> + *
> + * Attempt to get a valid hugepage size from /sys/kernel/mm/hugepages/ by
> + * reading directory contents
> + * Will fail (return -1) if the directory could not be opened or no valid
> + * page sizes are available. Will return the biggest hugepage size on
> + * success.
> + *
> + */
> +static long
> +InternalGetHugepageSize()
> +{
> + struct dirent *ent;
> + DIR *dir = opendir(HUGE_PAGE_INFO_DIR);
> + long smallest_size = -1, size;
> + char *ptr;
> +
> + if (dir == NULL)
> + {
> + ereport(huge_tlb_pages == HUGE_TLB_TRY ? DEBUG1 : WARNING,
> + (errmsg("Could not open directory %s: %s", HUGE_PAGE_INFO_DIR, strerror(errno)),
> + errcontext("while checking hugepage size")));
> + return -1;
> + }
> +
> + /*
> + * Linux supports multiple hugepage sizes if the hardware
> + * supports it; for each possible size there will be a
> + * directory in /sys/kernel/mm/hugepages consisting of the
> + * string hugepages- and the size of the page, e.g. on x86_64:
> + * hugepages-2048kB
> + */
> + while((ent = readdir(dir)) != NULL)
> + {
> + if (strncmp(ent->d_name, "hugepages-", 10) == 0)
> + {
> + size = strtol(ent->d_name + 10, &ptr, 10);
> + if (ptr == NULL)
> + {
> + continue;
> + }
> +
> + if (strcmp(ptr, "kB") == 0)
> + {
> + size *= 1024;
> + }
> +
> + if ((smallest_size == -1 || size < smallest_size)
> + && InternalGetFreeHugepagesCount(ent->d_name) > 0)
> + {
> + smallest_size = size;
> + }
> + }
> + }
> +
> + closedir(dir);
> +
> + if (smallest_size == -1)
> + {
> + ereport(huge_tlb_pages == HUGE_TLB_TRY ? DEBUG1 : WARNING,
> + (errmsg("Could not find a valid hugepage size"),
> + errhint("This error usually means that either CONFIG_HUGETLB_PAGE "
> + "is not in kernel or that your architecture does not "
> + "support hugepages or you did not configure hugepages")));
> + }
> +
> + return smallest_size;
> +}
> +#endif
> +
> /*
> * PGSharedMemoryCreate
> *
> @@ -391,7 +556,17 @@ PGSharedMemoryCreate(Size size, bool makePrivate, int port)
> */
> #ifndef EXEC_BACKEND
> {
> +#ifdef MAP_HUGETLB
> + long pagesize = 0;
> +
> + if (huge_tlb_pages == HUGE_TLB_ON || huge_tlb_pages == HUGE_TLB_TRY)
> + pagesize = InternalGetHugepageSize();
> +
> + if (pagesize <= 0)
> + pagesize = sysconf(_SC_PAGE_SIZE);
> +#else
> long pagesize = sysconf(_SC_PAGE_SIZE);
> +#endif
>
> /*
> * Ensure request size is a multiple of pagesize.
> @@ -410,8 +585,22 @@ PGSharedMemoryCreate(Size size, bool makePrivate, int port)
> * to be false, we might need to add a run-time test here and do this
> * only if the running kernel supports it.
> */
> - AnonymousShmem = mmap(NULL, size, PROT_READ|PROT_WRITE, PG_MMAP_FLAGS,
> - -1, 0);
> +
> + if (huge_tlb_pages == HUGE_TLB_ON || huge_tlb_pages == HUGE_TLB_TRY)
> + {
> + AnonymousShmem = mmap(PG_HUGETLB_BASE_ADDR, size, PROT_READ|PROT_WRITE,
> + PG_MMAP_FLAGS|PG_MAP_HUGETLB, -1, 0);
> +
> + elog(DEBUG3, "mmap() tried with MAP_HUGEPAGE: %p", AnonymousShmem);
> + }
> +
> + if ((AnonymousShmem == MAP_FAILED && huge_tlb_pages == HUGE_TLB_TRY)
> + || huge_tlb_pages == HUGE_TLB_OFF)
> + {
> + AnonymousShmem = mmap(NULL, size, PROT_READ|PROT_WRITE, PG_MMAP_FLAGS,
> + -1, 0);
> + }
> +
> if (AnonymousShmem == MAP_FAILED)
> ereport(FATAL,
> (errmsg("could not map anonymous shared memory: %m"),
> diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
> index 745e7be..28b6191 100644
> --- a/src/backend/utils/misc/guc.c
> +++ b/src/backend/utils/misc/guc.c
> @@ -22,6 +22,7 @@
> #include <limits.h>
> #include <unistd.h>
> #include <sys/stat.h>
> +#include <sys/mman.h>
> #ifdef HAVE_SYSLOG
> #include <syslog.h>
> #endif
> @@ -389,6 +390,22 @@ static const struct config_enum_entry synchronous_commit_options[] = {
> };
>
> /*
> + * huge_tlb_pages may be on|off|try, where try is the default
> + * on: try to mmap() with MAP_HUGETLB and fail when mmap() fails
> + * off: do not try tp mmap() with MAP_HUGETLB
> + * try: try to mmap() with MAP_HUGETLB and fallback to mmap()
> + * w/o MAP_HUGETLB
> + */
> +static const struct config_enum_entry huge_tlb_options[] = {
> +#ifdef MAP_HUGETLB
> + {"on", HUGE_TLB_ON, false},
> + {"try", HUGE_TLB_TRY, false},
> +#endif
> + {"off", HUGE_TLB_OFF, false},
> + {NULL, 0, false}
> +};
> +
> +/*
> * Options for enum values stored in other modules
> */
> extern const struct config_enum_entry wal_level_options[];
> @@ -447,6 +464,12 @@ int tcp_keepalives_idle;
> int tcp_keepalives_interval;
> int tcp_keepalives_count;
>
> +#ifdef MAP_HUGETLB
> +int huge_tlb_pages = HUGE_TLB_TRY;
> +#else
> +int huge_tlb_pages = HUGE_TLB_OFF;
> +#endif
> +
> /*
> * These variables are all dummies that don't do anything, except in some
> * cases provide the value for SHOW to display. The real state is elsewhere
> @@ -3301,6 +3324,26 @@ static struct config_enum ConfigureNamesEnum[] =
> NULL, NULL, NULL
> },
>
> + {
> + {"huge_tlb_pages",
> +#ifdef MAP_HUGETLB
> + PGC_SUSET,
> +#else
> + PGC_INTERNAL,
> +#endif
> + RESOURCES_MEM,
> + gettext_noop("Enable/disable the use of the hugepages feature"),
> + NULL
> + },
> + &huge_tlb_pages,
> +#ifdef MAP_HUGETLB
> + HUGE_TLB_TRY,
> +#else
> + HUGE_TLB_OFF,
> +#endif
> + huge_tlb_options,
> + NULL, NULL, NULL
> + },
>
> /* End-of-list marker */
> {
> diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
> index eeb9b82..e5bafec 100644
> --- a/src/backend/utils/misc/postgresql.conf.sample
> +++ b/src/backend/utils/misc/postgresql.conf.sample
> @@ -113,6 +113,7 @@
>
> #shared_buffers = 32MB # min 128kB
> # (change requires restart)
> +#huge_tlb_pages = try # try to map memory with MAP_HUGETLB (on, off, try)
> #temp_buffers = 8MB # min 800kB
> #max_prepared_transactions = 0 # zero disables the feature
> # (change requires restart)
> diff --git a/src/include/utils/guc.h b/src/include/utils/guc.h
> index 06f797c..17f5870 100644
> --- a/src/include/utils/guc.h
> +++ b/src/include/utils/guc.h
> @@ -230,6 +230,24 @@ extern int tcp_keepalives_idle;
> extern int tcp_keepalives_interval;
> extern int tcp_keepalives_count;
>
> +
> +/*
> + * Possible values for huge_tlb_pages; default is HUGE_TLB_TRY
> + */
> +typedef enum
> +{
> + HUGE_TLB_OFF,
> + HUGE_TLB_ON,
> + HUGE_TLB_TRY
> +} HugeTlbType;
> +
> +
> +/*
> + * configure the use of huge TLB pages
> + */
> +extern int huge_tlb_pages;
> +
> +
> /*
> * Functions exported by guc.c
> */

--
Bruce Momjian <bruce(at)momjian(dot)us> http://momjian.us
EnterpriseDB http://enterprisedb.com

+ It's impossible for everything to be true. +