Implement opt.metadata_thp

This option enables transparent huge page for base allocators (require
MADV_HUGEPAGE support).

configure.ac

@@ -1824,6 +1824,9 @@ if test "x${je_cv_madvise}" = "xyes" ; then
 	madvise((void *)0, 0, MADV_HUGEPAGE);
 	madvise((void *)0, 0, MADV_NOHUGEPAGE);
 ], [je_cv_thp])
+  if test "x${je_cv_thp}" = "xyes" ; then
+    AC_DEFINE([JEMALLOC_HAVE_MADVISE_HUGE], [ ])
+  fi
 fi
 
 dnl Enable transparent huge page support by default.

doc/jemalloc.xml.in

@@ -916,6 +916,18 @@ mallctl("arena." STRINGIFY(MALLCTL_ARENAS_ALL) ".decay",
         </para></listitem>
       </varlistentry>
 
+      <varlistentry id="opt.metadata_thp">
+        <term>
+          <mallctl>opt.metadata_thp</mallctl>
+          (<type>bool</type>)
+          <literal>r-</literal>
+        </term>
+        <listitem><para>If true, allow jemalloc to use transparent huge page
+        (THP) for internal metadata (see <link
+        linkend="stats.metadata">stats.metadata</link> for details).  This
+        option is disabled by default.</para></listitem>
+      </varlistentry>
+
       <varlistentry id="opt.retain">
         <term>
           <mallctl>opt.retain</mallctl>
@@ -2187,7 +2199,10 @@ struct extent_hooks_s {
         metadata structures (see <link
         linkend="stats.arenas.i.base"><mallctl>stats.arenas.&lt;i&gt;.base</mallctl></link>)
         and internal allocations (see <link
-        linkend="stats.arenas.i.internal"><mallctl>stats.arenas.&lt;i&gt;.internal</mallctl></link>).</para></listitem>
+        linkend="stats.arenas.i.internal"><mallctl>stats.arenas.&lt;i&gt;.internal</mallctl></link>).
+        Transparent huge page (enabled with <link
+        linkend="opt.metadata_thp">opt.metadata_thp</link>) usage is not
+        considered.</para></listitem>
       </varlistentry>
 
       <varlistentry id="stats.resident">

include/jemalloc/internal/base_externs.h

@@ -1,6 +1,8 @@
 #ifndef JEMALLOC_INTERNAL_BASE_EXTERNS_H
 #define JEMALLOC_INTERNAL_BASE_EXTERNS_H
 
+extern bool opt_metadata_thp;
+
 base_t *b0get(void);
 base_t *base_new(tsdn_t *tsdn, unsigned ind, extent_hooks_t *extent_hooks);
 void base_delete(tsdn_t *tsdn, base_t *base);

include/jemalloc/internal/base_types.h

@@ -4,4 +4,6 @@
 typedef struct base_block_s base_block_t;
 typedef struct base_s base_t;
 
+#define METADATA_THP_DEFAULT false
+
 #endif /* JEMALLOC_INTERNAL_BASE_TYPES_H */

include/jemalloc/internal/jemalloc_internal_defs.h.in

@@ -260,6 +260,12 @@
 /* Defined if madvise(2) is available. */
 #undef JEMALLOC_HAVE_MADVISE
 
+/*
+ * Defined if transparent huge pages are supported via the MADV_[NO]HUGEPAGE
+ * arguments to madvise(2).
+ */
+#undef JEMALLOC_HAVE_MADVISE_HUGE
+
 /*
  * Methods for purging unused pages differ between operating systems.
  *

include/jemalloc/internal/pages.h

@@ -58,6 +58,9 @@ static const bool pages_can_purge_forced =
 #endif
     ;
 
+/* Whether transparent huge page state is "madvise". */
+extern bool thp_state_madvise;
+
 void *pages_map(void *addr, size_t size, size_t alignment, bool *commit);
 void pages_unmap(void *addr, size_t size);
 bool pages_commit(void *addr, size_t size);

src/base.c

@@ -10,7 +10,9 @@
 /******************************************************************************/
 /* Data. */
 
-static base_t	*b0;
+static base_t *b0;
+
+bool opt_metadata_thp = METADATA_THP_DEFAULT;
 
 /******************************************************************************/
 
@@ -20,19 +22,26 @@ base_map(tsdn_t *tsdn, extent_hooks_t *extent_hooks, unsigned ind, size_t size)
 	bool zero = true;
 	bool commit = true;
 
+	/* We use hugepage sizes regardless of opt_metadata_thp. */
 	assert(size == HUGEPAGE_CEILING(size));
-
+	size_t alignment = opt_metadata_thp ? HUGEPAGE : PAGE;
 	if (extent_hooks == &extent_hooks_default) {
-		addr = extent_alloc_mmap(NULL, size, PAGE, &zero, &commit);
+		addr = extent_alloc_mmap(NULL, size, alignment, &zero, &commit);
 	} else {
 		/* No arena context as we are creating new arenas. */
 		tsd_t *tsd = tsdn_null(tsdn) ? tsd_fetch() : tsdn_tsd(tsdn);
 		pre_reentrancy(tsd, NULL);
-		addr = extent_hooks->alloc(extent_hooks, NULL, size, PAGE,
+		addr = extent_hooks->alloc(extent_hooks, NULL, size, alignment,
 		    &zero, &commit, ind);
 		post_reentrancy(tsd);
 	}
 
+	if (addr != NULL && opt_metadata_thp && thp_state_madvise) {
+		assert(((uintptr_t)addr & HUGEPAGE_MASK) == 0 &&
+		    (size & HUGEPAGE_MASK) == 0);
+		pages_huge(addr, size);
+	}
+
 	return addr;
 }
 
@@ -51,16 +60,16 @@ base_unmap(tsdn_t *tsdn, extent_hooks_t *extent_hooks, unsigned ind, void *addr,
 	 */
 	if (extent_hooks == &extent_hooks_default) {
 		if (!extent_dalloc_mmap(addr, size)) {
-			return;
+			goto label_done;
 		}
 		if (!pages_decommit(addr, size)) {
-			return;
+			goto label_done;
 		}
 		if (!pages_purge_forced(addr, size)) {
-			return;
+			goto label_done;
 		}
 		if (!pages_purge_lazy(addr, size)) {
-			return;
+			goto label_done;
 		}
 		/* Nothing worked.  This should never happen. */
 		not_reached();
@@ -70,27 +79,33 @@ base_unmap(tsdn_t *tsdn, extent_hooks_t *extent_hooks, unsigned ind, void *addr,
 		if (extent_hooks->dalloc != NULL &&
 		    !extent_hooks->dalloc(extent_hooks, addr, size, true,
 		    ind)) {
-			goto label_done;
+			goto label_post_reentrancy;
 		}
 		if (extent_hooks->decommit != NULL &&
 		    !extent_hooks->decommit(extent_hooks, addr, size, 0, size,
 		    ind)) {
-			goto label_done;
+			goto label_post_reentrancy;
 		}
 		if (extent_hooks->purge_forced != NULL &&
 		    !extent_hooks->purge_forced(extent_hooks, addr, size, 0,
 		    size, ind)) {
-			goto label_done;
+			goto label_post_reentrancy;
 		}
 		if (extent_hooks->purge_lazy != NULL &&
 		    !extent_hooks->purge_lazy(extent_hooks, addr, size, 0, size,
 		    ind)) {
-			goto label_done;
+			goto label_post_reentrancy;
 		}
 		/* Nothing worked.  That's the application's problem. */
-	label_done:
+	label_post_reentrancy:
 		post_reentrancy(tsd);
-		return;
+	}
+label_done:
+	if (opt_metadata_thp && thp_state_madvise) {
+		/* Set NOHUGEPAGE after unmap to avoid kernel defrag. */
+		assert(((uintptr_t)addr & HUGEPAGE_MASK) == 0 &&
+		    (size & HUGEPAGE_MASK) == 0);
+		pages_nohuge(addr, size);
 	}
 }
 

src/ctl.c

@@ -80,6 +80,7 @@ CTL_PROTO(config_utrace)
 CTL_PROTO(config_xmalloc)
 CTL_PROTO(opt_abort)
 CTL_PROTO(opt_abort_conf)
+CTL_PROTO(opt_metadata_thp)
 CTL_PROTO(opt_retain)
 CTL_PROTO(opt_dss)
 CTL_PROTO(opt_narenas)
@@ -274,6 +275,7 @@ static const ctl_named_node_t	config_node[] = {
 static const ctl_named_node_t opt_node[] = {
 	{NAME("abort"),		CTL(opt_abort)},
 	{NAME("abort_conf"),	CTL(opt_abort_conf)},
+	{NAME("metadata_thp"),	CTL(opt_metadata_thp)},
 	{NAME("retain"),	CTL(opt_retain)},
 	{NAME("dss"),		CTL(opt_dss)},
 	{NAME("narenas"),	CTL(opt_narenas)},
@@ -1568,6 +1570,7 @@ CTL_RO_CONFIG_GEN(config_xmalloc, bool)
 
 CTL_RO_NL_GEN(opt_abort, opt_abort, bool)
 CTL_RO_NL_GEN(opt_abort_conf, opt_abort_conf, bool)
+CTL_RO_NL_GEN(opt_metadata_thp, opt_metadata_thp, bool)
 CTL_RO_NL_GEN(opt_retain, opt_retain, bool)
 CTL_RO_NL_GEN(opt_dss, opt_dss, const char *)
 CTL_RO_NL_GEN(opt_narenas, opt_narenas, unsigned)

src/jemalloc.c

@@ -1055,6 +1055,7 @@ malloc_conf_init(void) {
 			if (opt_abort_conf && had_conf_error) {
 				malloc_abort_invalid_conf();
 			}
+			CONF_HANDLE_BOOL(opt_metadata_thp, "metadata_thp")
 			CONF_HANDLE_BOOL(opt_retain, "retain")
 			if (strncmp("dss", k, klen) == 0) {
 				int i;

src/pages.c

@@ -25,6 +25,8 @@ static int	mmap_flags;
 #endif
 static bool	os_overcommits;
 
+bool thp_state_madvise;
+
 /******************************************************************************/
 /*
  * Function prototypes for static functions that are referenced prior to
@@ -291,7 +293,7 @@ pages_huge(void *addr, size_t size) {
 	assert(HUGEPAGE_ADDR2BASE(addr) == addr);
 	assert(HUGEPAGE_CEILING(size) == size);
 
-#ifdef JEMALLOC_THP
+#ifdef JEMALLOC_HAVE_MADVISE_HUGE
 	return (madvise(addr, size, MADV_HUGEPAGE) != 0);
 #else
 	return true;
@@ -303,7 +305,7 @@ pages_nohuge(void *addr, size_t size) {
 	assert(HUGEPAGE_ADDR2BASE(addr) == addr);
 	assert(HUGEPAGE_CEILING(size) == size);
 
-#ifdef JEMALLOC_THP
+#ifdef JEMALLOC_HAVE_MADVISE_HUGE
 	return (madvise(addr, size, MADV_NOHUGEPAGE) != 0);
 #else
 	return false;
@@ -413,6 +415,51 @@ os_overcommits_proc(void) {
 }
 #endif
 
+static void
+init_thp_state(void) {
+#ifndef JEMALLOC_HAVE_MADVISE_HUGE
+	if (opt_metadata_thp && opt_abort) {
+		malloc_write("<jemalloc>: no MADV_HUGEPAGE support\n");
+		abort();
+	}
+	goto label_error;
+#endif
+	static const char madvise_state[] = "always [madvise] never\n";
+	char buf[sizeof(madvise_state)];
+
+#if defined(JEMALLOC_USE_SYSCALL) && defined(SYS_open)
+	int fd = (int)syscall(SYS_open,
+	    "/sys/kernel/mm/transparent_hugepage/enabled", O_RDONLY);
+#else
+	int fd = open("/sys/kernel/mm/transparent_hugepage/enabled", O_RDONLY);
+#endif
+	if (fd == -1) {
+		goto label_error;
+	}
+
+#if defined(JEMALLOC_USE_SYSCALL) && defined(SYS_read)
+	ssize_t nread = (ssize_t)syscall(SYS_read, fd, &buf, sizeof(buf));
+#else
+	ssize_t nread = read(fd, &buf, sizeof(buf));
+#endif
+
+#if defined(JEMALLOC_USE_SYSCALL) && defined(SYS_close)
+	syscall(SYS_close, fd);
+#else
+	close(fd);
+#endif
+
+	if (nread < 1) {
+		goto label_error;
+	}
+	if (strncmp(buf, madvise_state, (size_t)nread) == 0) {
+		thp_state_madvise = true;
+		return;
+	}
+label_error:
+	thp_state_madvise = false;
+}
+
 bool
 pages_boot(void) {
 	os_page = os_page_detect();
@@ -441,5 +488,7 @@ pages_boot(void) {
 	os_overcommits = false;
 #endif
 
+	init_thp_state();
+
 	return false;
 }

src/stats.c

@@ -802,6 +802,7 @@ stats_general_print(void (*write_cb)(void *, const char *), void *cbopaque,
 	}
 	OPT_WRITE_BOOL(abort, ",")
 	OPT_WRITE_BOOL(abort_conf, ",")
+	OPT_WRITE_BOOL(metadata_thp, ",")
 	OPT_WRITE_BOOL(retain, ",")
 	OPT_WRITE_CHAR_P(dss, ",")
 	OPT_WRITE_UNSIGNED(narenas, ",")

test/unit/mallctl.c

@@ -157,6 +157,7 @@ TEST_BEGIN(test_mallctl_opt) {
 } while (0)
 
 	TEST_MALLCTL_OPT(bool, abort, always);
+	TEST_MALLCTL_OPT(bool, metadata_thp, always);
 	TEST_MALLCTL_OPT(bool, retain, always);
 	TEST_MALLCTL_OPT(const char *, dss, always);
 	TEST_MALLCTL_OPT(unsigned, narenas, always);