At 2013-10-30 11:04:36 -0400, t...@sss.pgh.pa.us wrote:
>
> > As a compromise, perhaps we can unconditionally round the size up to be
> > a multiple of 2MB? […]
>
> That sounds reasonably painless to me.
Here's a patch that does that and adds a DEBUG1 log message when we try
with MAP_HUGETLB and fail and fallback to ordinary mmap.
-- Abhijit
diff --git a/src/backend/utils/misc/postgresql.conf.sample b/src/backend/utils/misc/postgresql.conf.sample
index 707edf1..7d9d0a8 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/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index 77a9303..e4ded7a 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -1052,6 +1052,49 @@ 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>try</literal> (the default), <literal>on</literal>,
+ and <literal>off</literal>.
+ </para>
+
+ <para>
+ At present, this feature is supported only on Linux. The setting
+ is ignored on other systems.
+ </para>
+
+ <para>
+ The use of huge TLB pages results in smaller page tables and
+ less CPU time spent on memory management. For more details, see
+ <ulink url="https://www.kernel.org/doc/Documentation/vm/hugetlbpage.txt";>hugepages.txt
+ </ulink> in the Linux kernel documentation.
+ </para>
+
+ <para>
+ With <varname>huge_tlb_pages</varname> set to <literal>try</literal>,
+ the server will try to use huge pages, but fall back to using
+ normal allocation if the first attempt fails.
+ </para>
+
+ <para>
+ With <varname>huge_tlb_pages</varname> set to <literal>on</literal>,
+ the server will try to use huge pages, and treat failure as a
+ FATAL error.
+ </para>
+
+ <para>
+ With <varname>huge_tlb_pages</varname> set to <literal>off</literal>,
+ the server will not try to use huge pages.
+ </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 b604407..fc0d74b 100644
--- a/src/backend/port/sysv_shmem.c
+++ b/src/backend/port/sysv_shmem.c
@@ -32,6 +32,7 @@
#include "portability/mem.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) */
@@ -367,14 +368,31 @@ PGSharedMemoryCreate(Size size, bool makePrivate, int port)
*/
#ifndef EXEC_BACKEND
{
- long pagesize = sysconf(_SC_PAGE_SIZE);
+ int flags = PG_MMAP_FLAGS;
+ long pagesize = 2*1024*1024;
+
+#ifdef MAP_HUGETLB
+ if (huge_tlb_pages == HUGE_TLB_ON || huge_tlb_pages == HUGE_TLB_TRY)
+ {
+ flags |= MAP_HUGETLB;
+ }
+#endif
/*
* Ensure request size is a multiple of pagesize.
*
- * pagesize will, for practical purposes, always be a power of two.
- * But just in case it isn't, we do it this way instead of using
- * TYPEALIGN().
+ * By doing this ourselves, we maximise the chances of being
+ * able to use huge TLB pages even on kernels that do not round
+ * up the request size correctly, for example due to this bug:
+ * https://bugzilla.kernel.org/show_bug.cgi?id=56881
+ *
+ * The default value of 2MB for pagesize is chosen based on the
+ * most common supported huge page size. Rounding up to a larger
+ * value (e.g. 16MB) would use even larger pages if the hardware
+ * supported them, but would potentially waste more space.
+ *
+ * We round up by hand instead of using TYPEALIGN(), but for all
+ * practical purposes, pagesize will always be a power of two.
*/
if (pagesize > 0 && size % pagesize != 0)
size += pagesize - (size % pagesize);
@@ -386,8 +404,21 @@ PGSharedMemoryCreate(Size size, bool makePrivate, int port)
* out 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);
+
+ AnonymousShmem = mmap(NULL, size, PROT_READ|PROT_WRITE, flags, -1, 0);
+
+#ifdef MAP_HUGETLB
+ if (huge_tlb_pages == HUGE_TLB_TRY && AnonymousShmem == MAP_FAILED)
+ {
+ elog(DEBUG1, "mmap(%lu) with MAP_HUGETLB failed with errno=%d; "
+ "trying without", (uint64)size, errno);
+
+ flags &= ~MAP_HUGETLB;
+ AnonymousShmem = mmap(NULL, size, PROT_READ|PROT_WRITE, flags,
+ -1, 0);
+ }
+#endif
+
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 dfc6704..8faafb4 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -381,6 +381,26 @@ 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[] = {
+ {"off", HUGE_TLB_OFF, false},
+ {"on", HUGE_TLB_ON, false},
+ {"try", HUGE_TLB_TRY, false},
+ {"true", HUGE_TLB_ON, true},
+ {"false", HUGE_TLB_OFF, true},
+ {"yes", HUGE_TLB_ON, true},
+ {"no", HUGE_TLB_OFF, true},
+ {"1", HUGE_TLB_ON, true},
+ {"0", HUGE_TLB_OFF, true},
+ {NULL, 0, false}
+};
+
+/*
* Options for enum values stored in other modules
*/
extern const struct config_enum_entry wal_level_options[];
@@ -440,6 +460,8 @@ int tcp_keepalives_idle;
int tcp_keepalives_interval;
int tcp_keepalives_count;
+int huge_tlb_pages = HUGE_TLB_TRY;
+
/*
* 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
@@ -3377,6 +3399,18 @@ static struct config_enum ConfigureNamesEnum[] =
NULL, NULL, NULL
},
+ {
+ {"huge_tlb_pages",
+ PGC_POSTMASTER,
+ RESOURCES_MEM,
+ gettext_noop("Enable/disable the use of huge TLB pages on Linux"),
+ NULL
+ },
+ &huge_tlb_pages,
+ HUGE_TLB_TRY,
+ huge_tlb_options,
+ NULL, NULL, NULL
+ },
/* End-of-list marker */
{
diff --git a/src/include/utils/guc.h b/src/include/utils/guc.h
index 3e981b3..8011e88 100644
--- a/src/include/utils/guc.h
+++ b/src/include/utils/guc.h
@@ -235,6 +235,24 @@ 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
*/
diff --git a/src/include/pg_config.h.in b/src/include/pg_config.h.in
index 0250e39..051b6cf 100644
--- a/src/include/pg_config.h.in
+++ b/src/include/pg_config.h.in
@@ -532,6 +532,9 @@
/* Define to 1 if you have the <sys/ipc.h> header file. */
#undef HAVE_SYS_IPC_H
+/* Define to 1 if you have the <sys/mman.h> header file. */
+#undef HAVE_SYS_MMAN_H
+
/* Define to 1 if you have the <sys/poll.h> header file. */
#undef HAVE_SYS_POLL_H
diff --git a/configure b/configure
index c20afde..67bc57f 100755
--- a/configure
+++ b/configure
@@ -10524,7 +10524,7 @@ done
-for ac_header in crypt.h dld.h fp_class.h getopt.h ieeefp.h ifaddrs.h langinfo.h poll.h pwd.h sys/ioctl.h sys/ipc.h sys/poll.h sys/pstat.h sys/resource.h sys/select.h sys/sem.h sys/shm.h sys/socket.h sys/sockio.h sys/tas.h sys/time.h sys/un.h termios.h ucred.h utime.h wchar.h wctype.h
+for ac_header in crypt.h dld.h fp_class.h getopt.h ieeefp.h ifaddrs.h langinfo.h poll.h pwd.h sys/ioctl.h sys/ipc.h sys/mman.h sys/poll.h sys/pstat.h sys/resource.h sys/select.h sys/sem.h sys/shm.h sys/socket.h sys/sockio.h sys/tas.h sys/time.h sys/un.h termios.h ucred.h utime.h wchar.h wctype.h
do
as_ac_Header=`$as_echo "ac_cv_header_$ac_header" | $as_tr_sh`
if { as_var=$as_ac_Header; eval "test \"\${$as_var+set}\" = set"; }; then
diff --git a/configure.in b/configure.in
index d2bab32..b755202 100644
--- a/configure.in
+++ b/configure.in
@@ -982,7 +982,7 @@ AC_SUBST(OSSP_UUID_LIBS)
##
dnl sys/socket.h is required by AC_FUNC_ACCEPT_ARGTYPES
-AC_CHECK_HEADERS([crypt.h dld.h fp_class.h getopt.h ieeefp.h ifaddrs.h langinfo.h poll.h pwd.h sys/ioctl.h sys/ipc.h sys/poll.h sys/pstat.h sys/resource.h sys/select.h sys/sem.h sys/shm.h sys/socket.h sys/sockio.h sys/tas.h sys/time.h sys/un.h termios.h ucred.h utime.h wchar.h wctype.h])
+AC_CHECK_HEADERS([crypt.h dld.h fp_class.h getopt.h ieeefp.h ifaddrs.h langinfo.h poll.h pwd.h sys/ioctl.h sys/ipc.h sys/mman.h sys/poll.h sys/pstat.h sys/resource.h sys/select.h sys/sem.h sys/shm.h sys/socket.h sys/sockio.h sys/tas.h sys/time.h sys/un.h termios.h ucred.h utime.h wchar.h wctype.h])
# On BSD, test for net/if.h will fail unless sys/socket.h
# is included first.
--
Sent via pgsql-hackers mailing list (pgsql-hackers@postgresql.org)
To make changes to your subscription:
http://www.postgresql.org/mailpref/pgsql-hackers
