Core/Dependencies: Upgrade to jemalloc-3.3.1 (Also fixes a few weird crashes), closes #7852

Note: This adds support for using jemalloc on FreeBSD, Mac OS X Lion and a few other platforms (not enabled by default)
Please read dep/jemalloc/ChangeLog for further information (Changed from 2.1.1 and up apply)

1 files changed, 138 insertions, 366 deletions

diff --git a/dep/jemalloc/src/stats.c b/dep/jemalloc/src/stats.c
index 3dfe0d232a6..43f87af6700 100644
--- a/dep/jemalloc/src/stats.c
+++ b/dep/jemalloc/src/stats.c
@@ -39,136 +39,40 @@
 
 bool	opt_stats_print = false;
 
+size_t	stats_cactive = 0;
+
 /******************************************************************************/
 /* Function prototypes for non-inline static functions. */
 
-#ifdef JEMALLOC_STATS
-static void	malloc_vcprintf(void (*write_cb)(void *, const char *),
-    void *cbopaque, const char *format, va_list ap);
 static void	stats_arena_bins_print(void (*write_cb)(void *, const char *),
     void *cbopaque, unsigned i);
 static void	stats_arena_lruns_print(void (*write_cb)(void *, const char *),
     void *cbopaque, unsigned i);
 static void	stats_arena_print(void (*write_cb)(void *, const char *),
-    void *cbopaque, unsigned i);
-#endif
+    void *cbopaque, unsigned i, bool bins, bool large);
 
 /******************************************************************************/
 
-/*
- * We don't want to depend on vsnprintf() for production builds, since that can
- * cause unnecessary bloat for static binaries.  u2s() provides minimal integer
- * printing functionality, so that malloc_printf() use can be limited to
- * JEMALLOC_STATS code.
- */
-char *
-u2s(uint64_t x, unsigned base, char *s)
-{
-	unsigned i;
-
-	i = UMAX2S_BUFSIZE - 1;
-	s[i] = '\0';
-	switch (base) {
-	case 10:
-		do {
-			i--;
-			s[i] = "0123456789"[x % (uint64_t)10];
-			x /= (uint64_t)10;
-		} while (x > 0);
-		break;
-	case 16:
-		do {
-			i--;
-			s[i] = "0123456789abcdef"[x & 0xf];
-			x >>= 4;
-		} while (x > 0);
-		break;
-	default:
-		do {
-			i--;
-			s[i] = "0123456789abcdefghijklmnopqrstuvwxyz"[x %
-			    (uint64_t)base];
-			x /= (uint64_t)base;
-		} while (x > 0);
-	}
-
-	return (&s[i]);
-}
-
-#ifdef JEMALLOC_STATS
-static void
-malloc_vcprintf(void (*write_cb)(void *, const char *), void *cbopaque,
-    const char *format, va_list ap)
-{
-	char buf[4096];
-
-	if (write_cb == NULL) {
-		/*
-		 * The caller did not provide an alternate write_cb callback
-		 * function, so use the default one.  malloc_write() is an
-		 * inline function, so use malloc_message() directly here.
-		 */
-		write_cb = JEMALLOC_P(malloc_message);
-		cbopaque = NULL;
-	}
-
-	vsnprintf(buf, sizeof(buf), format, ap);
-	write_cb(cbopaque, buf);
-}
-
-/*
- * Print to a callback function in such a way as to (hopefully) avoid memory
- * allocation.
- */
-JEMALLOC_ATTR(format(printf, 3, 4))
-void
-malloc_cprintf(void (*write_cb)(void *, const char *), void *cbopaque,
-    const char *format, ...)
-{
-	va_list ap;
-
-	va_start(ap, format);
-	malloc_vcprintf(write_cb, cbopaque, format, ap);
-	va_end(ap);
-}
-
-/*
- * Print to stderr in such a way as to (hopefully) avoid memory allocation.
- */
-JEMALLOC_ATTR(format(printf, 1, 2))
-void
-malloc_printf(const char *format, ...)
-{
-	va_list ap;
-
-	va_start(ap, format);
-	malloc_vcprintf(NULL, NULL, format, ap);
-	va_end(ap);
-}
-#endif
-
-#ifdef JEMALLOC_STATS
 static void
 stats_arena_bins_print(void (*write_cb)(void *, const char *), void *cbopaque,
     unsigned i)
 {
-	size_t pagesize;
+	size_t page;
 	bool config_tcache;
 	unsigned nbins, j, gap_start;
 
-	CTL_GET("arenas.pagesize", &pagesize, size_t);
+	CTL_GET("arenas.page", &page, size_t);
 
 	CTL_GET("config.tcache", &config_tcache, bool);
 	if (config_tcache) {
 		malloc_cprintf(write_cb, cbopaque,
-		    "bins:     bin    size regs pgs    allocated      nmalloc"
+		    "bins:     bin  size regs pgs    allocated      nmalloc"
 		    "      ndalloc    nrequests       nfills     nflushes"
-		    "      newruns       reruns      maxruns      curruns\n");
+		    "      newruns       reruns      curruns\n");
 	} else {
 		malloc_cprintf(write_cb, cbopaque,
-		    "bins:     bin    size regs pgs    allocated      nmalloc"
-		    "      ndalloc      newruns       reruns      maxruns"
-		    "      curruns\n");
+		    "bins:     bin  size regs pgs    allocated      nmalloc"
+		    "      ndalloc      newruns       reruns      curruns\n");
 	}
 	CTL_GET("arenas.nbins", &nbins, unsigned);
 	for (j = 0, gap_start = UINT_MAX; j < nbins; j++) {
@@ -179,12 +83,11 @@ stats_arena_bins_print(void (*write_cb)(void *, const char *), void *cbopaque,
 			if (gap_start == UINT_MAX)
 				gap_start = j;
 		} else {
-			unsigned ntbins_, nqbins, ncbins, nsbins;
 			size_t reg_size, run_size, allocated;
 			uint32_t nregs;
 			uint64_t nmalloc, ndalloc, nrequests, nfills, nflushes;
 			uint64_t reruns;
-			size_t highruns, curruns;
+			size_t curruns;
 
 			if (gap_start != UINT_MAX) {
 				if (j > gap_start + 1) {
@@ -199,10 +102,6 @@ stats_arena_bins_print(void (*write_cb)(void *, const char *), void *cbopaque,
 				}
 				gap_start = UINT_MAX;
 			}
-			CTL_GET("arenas.ntbins", &ntbins_, unsigned);
-			CTL_GET("arenas.nqbins", &nqbins, unsigned);
-			CTL_GET("arenas.ncbins", &ncbins, unsigned);
-			CTL_GET("arenas.nsbins", &nsbins, unsigned);
 			CTL_J_GET("arenas.bin.0.size", &reg_size, size_t);
 			CTL_J_GET("arenas.bin.0.nregs", &nregs, uint32_t);
 			CTL_J_GET("arenas.bin.0.run_size", &run_size, size_t);
@@ -222,36 +121,25 @@ stats_arena_bins_print(void (*write_cb)(void *, const char *), void *cbopaque,
 			}
 			CTL_IJ_GET("stats.arenas.0.bins.0.nreruns", &reruns,
 			    uint64_t);
-			CTL_IJ_GET("stats.arenas.0.bins.0.highruns", &highruns,
-			    size_t);
 			CTL_IJ_GET("stats.arenas.0.bins.0.curruns", &curruns,
 			    size_t);
 			if (config_tcache) {
 				malloc_cprintf(write_cb, cbopaque,
-				    "%13u %1s %5zu %4u %3zu %12zu %12"PRIu64
+				    "%13u %5zu %4u %3zu %12zu %12"PRIu64
 				    " %12"PRIu64" %12"PRIu64" %12"PRIu64
 				    " %12"PRIu64" %12"PRIu64" %12"PRIu64
-				    " %12zu %12zu\n",
-				    j,
-				    j < ntbins_ ? "T" : j < ntbins_ + nqbins ?
-				    "Q" : j < ntbins_ + nqbins + ncbins ? "C" :
-				    "S",
-				    reg_size, nregs, run_size / pagesize,
+				    " %12zu\n",
+				    j, reg_size, nregs, run_size / page,
 				    allocated, nmalloc, ndalloc, nrequests,
-				    nfills, nflushes, nruns, reruns, highruns,
-				    curruns);
+				    nfills, nflushes, nruns, reruns, curruns);
 			} else {
 				malloc_cprintf(write_cb, cbopaque,
-				    "%13u %1s %5zu %4u %3zu %12zu %12"PRIu64
+				    "%13u %5zu %4u %3zu %12zu %12"PRIu64
 				    " %12"PRIu64" %12"PRIu64" %12"PRIu64
-				    " %12zu %12zu\n",
-				    j,
-				    j < ntbins_ ? "T" : j < ntbins_ + nqbins ?
-				    "Q" : j < ntbins_ + nqbins + ncbins ? "C" :
-				    "S",
-				    reg_size, nregs, run_size / pagesize,
+				    " %12zu\n",
+				    j, reg_size, nregs, run_size / page,
 				    allocated, nmalloc, ndalloc, nruns, reruns,
-				    highruns, curruns);
+				    curruns);
 			}
 		}
 	}
@@ -271,18 +159,18 @@ static void
 stats_arena_lruns_print(void (*write_cb)(void *, const char *), void *cbopaque,
     unsigned i)
 {
-	size_t pagesize, nlruns, j;
+	size_t page, nlruns, j;
 	ssize_t gap_start;
 
-	CTL_GET("arenas.pagesize", &pagesize, size_t);
+	CTL_GET("arenas.page", &page, size_t);
 
 	malloc_cprintf(write_cb, cbopaque,
 	    "large:   size pages      nmalloc      ndalloc    nrequests"
-	    "      maxruns      curruns\n");
+	    "      curruns\n");
 	CTL_GET("arenas.nlruns", &nlruns, size_t);
 	for (j = 0, gap_start = -1; j < nlruns; j++) {
 		uint64_t nmalloc, ndalloc, nrequests;
-		size_t run_size, highruns, curruns;
+		size_t run_size, curruns;
 
 		CTL_IJ_GET("stats.arenas.0.lruns.0.nmalloc", &nmalloc,
 		    uint64_t);
@@ -295,8 +183,6 @@ stats_arena_lruns_print(void (*write_cb)(void *, const char *), void *cbopaque,
 				gap_start = j;
 		} else {
 			CTL_J_GET("arenas.lrun.0.size", &run_size, size_t);
-			CTL_IJ_GET("stats.arenas.0.lruns.0.highruns", &highruns,
-			    size_t);
 			CTL_IJ_GET("stats.arenas.0.lruns.0.curruns", &curruns,
 			    size_t);
 			if (gap_start != -1) {
@@ -306,9 +192,9 @@ stats_arena_lruns_print(void (*write_cb)(void *, const char *), void *cbopaque,
 			}
 			malloc_cprintf(write_cb, cbopaque,
 			    "%13zu %5zu %12"PRIu64" %12"PRIu64" %12"PRIu64
-			    " %12zu %12zu\n",
-			    run_size, run_size / pagesize, nmalloc, ndalloc,
-			    nrequests, highruns, curruns);
+			    " %12zu\n",
+			    run_size, run_size / page, nmalloc, ndalloc,
+			    nrequests, curruns);
 		}
 	}
 	if (gap_start != -1)
@@ -317,17 +203,25 @@ stats_arena_lruns_print(void (*write_cb)(void *, const char *), void *cbopaque,
 
 static void
 stats_arena_print(void (*write_cb)(void *, const char *), void *cbopaque,
-    unsigned i)
+    unsigned i, bool bins, bool large)
 {
-	size_t pagesize, pactive, pdirty, mapped;
+	unsigned nthreads;
+	const char *dss;
+	size_t page, pactive, pdirty, mapped;
 	uint64_t npurge, nmadvise, purged;
 	size_t small_allocated;
 	uint64_t small_nmalloc, small_ndalloc, small_nrequests;
 	size_t large_allocated;
 	uint64_t large_nmalloc, large_ndalloc, large_nrequests;
 
-	CTL_GET("arenas.pagesize", &pagesize, size_t);
+	CTL_GET("arenas.page", &page, size_t);
 
+	CTL_I_GET("stats.arenas.0.nthreads", &nthreads, unsigned);
+	malloc_cprintf(write_cb, cbopaque,
+	    "assigned threads: %u\n", nthreads);
+	CTL_I_GET("stats.arenas.0.dss", &dss, const char *);
+	malloc_cprintf(write_cb, cbopaque, "dss allocation precedence: %s\n",
+	    dss);
 	CTL_I_GET("stats.arenas.0.pactive", &pactive, size_t);
 	CTL_I_GET("stats.arenas.0.pdirty", &pdirty, size_t);
 	CTL_I_GET("stats.arenas.0.npurge", &npurge, uint64_t);
@@ -361,15 +255,15 @@ stats_arena_print(void (*write_cb)(void *, const char *), void *cbopaque,
 	    small_nmalloc + large_nmalloc,
 	    small_ndalloc + large_ndalloc,
 	    small_nrequests + large_nrequests);
-	malloc_cprintf(write_cb, cbopaque, "active:  %12zu\n",
-	    pactive * pagesize );
+	malloc_cprintf(write_cb, cbopaque, "active:  %12zu\n", pactive * page);
 	CTL_I_GET("stats.arenas.0.mapped", &mapped, size_t);
 	malloc_cprintf(write_cb, cbopaque, "mapped:  %12zu\n", mapped);
 
-	stats_arena_bins_print(write_cb, cbopaque, i);
-	stats_arena_lruns_print(write_cb, cbopaque, i);
+	if (bins)
+		stats_arena_bins_print(write_cb, cbopaque, i);
+	if (large)
+		stats_arena_lruns_print(write_cb, cbopaque, i);
 }
-#endif
 
 void
 stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
@@ -378,7 +272,6 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 	int err;
 	uint64_t epoch;
 	size_t u64sz;
-	char s[UMAX2S_BUFSIZE];
 	bool general = true;
 	bool merged = true;
 	bool unmerged = true;
@@ -394,8 +287,7 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 	 * */
 	epoch = 1;
 	u64sz = sizeof(uint64_t);
-	err = JEMALLOC_P(mallctl)("epoch", &epoch, &u64sz, &epoch,
-	    sizeof(uint64_t));
+	err = je_mallctl("epoch", &epoch, &u64sz, &epoch, sizeof(uint64_t));
 	if (err != 0) {
 		if (err == EAGAIN) {
 			malloc_write("<jemalloc>: Memory allocation failure in "
@@ -407,42 +299,33 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 		abort();
 	}
 
-	if (write_cb == NULL) {
-		/*
-		 * The caller did not provide an alternate write_cb callback
-		 * function, so use the default one.  malloc_write() is an
-		 * inline function, so use malloc_message() directly here.
-		 */
-		write_cb = JEMALLOC_P(malloc_message);
-		cbopaque = NULL;
-	}
-
 	if (opts != NULL) {
 		unsigned i;
 
 		for (i = 0; opts[i] != '\0'; i++) {
 			switch (opts[i]) {
-				case 'g':
-					general = false;
-					break;
-				case 'm':
-					merged = false;
-					break;
-				case 'a':
-					unmerged = false;
-					break;
-				case 'b':
-					bins = false;
-					break;
-				case 'l':
-					large = false;
-					break;
-				default:;
+			case 'g':
+				general = false;
+				break;
+			case 'm':
+				merged = false;
+				break;
+			case 'a':
+				unmerged = false;
+				break;
+			case 'b':
+				bins = false;
+				break;
+			case 'l':
+				large = false;
+				break;
+			default:;
 			}
 		}
 	}
 
-	write_cb(cbopaque, "___ Begin jemalloc statistics ___\n");
+	malloc_cprintf(write_cb, cbopaque,
+	    "___ Begin jemalloc statistics ___\n");
 	if (general) {
 		int err;
 		const char *cpv;
@@ -457,256 +340,146 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 		cpsz = sizeof(const char *);
 
 		CTL_GET("version", &cpv, const char *);
-		write_cb(cbopaque, "Version: ");
-		write_cb(cbopaque, cpv);
-		write_cb(cbopaque, "\n");
+		malloc_cprintf(write_cb, cbopaque, "Version: %s\n", cpv);
 		CTL_GET("config.debug", &bv, bool);
-		write_cb(cbopaque, "Assertions ");
-		write_cb(cbopaque, bv ? "enabled" : "disabled");
-		write_cb(cbopaque, "\n");
+		malloc_cprintf(write_cb, cbopaque, "Assertions %s\n",
+		    bv ? "enabled" : "disabled");
 
 #define OPT_WRITE_BOOL(n)						\
-		if ((err = JEMALLOC_P(mallctl)("opt."#n, &bv, &bsz,	\
-		    NULL, 0)) == 0) {					\
-			write_cb(cbopaque, "  opt."#n": ");		\
-			write_cb(cbopaque, bv ? "true" : "false");	\
-			write_cb(cbopaque, "\n");			\
+		if ((err = je_mallctl("opt."#n, &bv, &bsz, NULL, 0))	\
+		    == 0) {						\
+			malloc_cprintf(write_cb, cbopaque,		\
+			    "  opt."#n": %s\n", bv ? "true" : "false");	\
 		}
 #define OPT_WRITE_SIZE_T(n)						\
-		if ((err = JEMALLOC_P(mallctl)("opt."#n, &sv, &ssz,	\
-		    NULL, 0)) == 0) {					\
-			write_cb(cbopaque, "  opt."#n": ");		\
-			write_cb(cbopaque, u2s(sv, 10, s));		\
-			write_cb(cbopaque, "\n");			\
+		if ((err = je_mallctl("opt."#n, &sv, &ssz, NULL, 0))	\
+		    == 0) {						\
+			malloc_cprintf(write_cb, cbopaque,		\
+			"  opt."#n": %zu\n", sv);			\
 		}
 #define OPT_WRITE_SSIZE_T(n)						\
-		if ((err = JEMALLOC_P(mallctl)("opt."#n, &ssv, &sssz,	\
-		    NULL, 0)) == 0) {					\
-			if (ssv >= 0) {					\
-				write_cb(cbopaque, "  opt."#n": ");	\
-				write_cb(cbopaque, u2s(ssv, 10, s));	\
-			} else {					\
-				write_cb(cbopaque, "  opt."#n": -");	\
-				write_cb(cbopaque, u2s(-ssv, 10, s));	\
-			}						\
-			write_cb(cbopaque, "\n");			\
+		if ((err = je_mallctl("opt."#n, &ssv, &sssz, NULL, 0))	\
+		    == 0) {						\
+			malloc_cprintf(write_cb, cbopaque,		\
+			    "  opt."#n": %zd\n", ssv);			\
 		}
 #define OPT_WRITE_CHAR_P(n)						\
-		if ((err = JEMALLOC_P(mallctl)("opt."#n, &cpv, &cpsz,	\
-		    NULL, 0)) == 0) {					\
-			write_cb(cbopaque, "  opt."#n": \"");		\
-			write_cb(cbopaque, cpv);			\
-			write_cb(cbopaque, "\"\n");			\
+		if ((err = je_mallctl("opt."#n, &cpv, &cpsz, NULL, 0))	\
+		    == 0) {						\
+			malloc_cprintf(write_cb, cbopaque,		\
+			    "  opt."#n": \"%s\"\n", cpv);		\
 		}
 
-		write_cb(cbopaque, "Run-time option settings:\n");
+		malloc_cprintf(write_cb, cbopaque,
+		    "Run-time option settings:\n");
 		OPT_WRITE_BOOL(abort)
-		OPT_WRITE_SIZE_T(lg_qspace_max)
-		OPT_WRITE_SIZE_T(lg_cspace_max)
 		OPT_WRITE_SIZE_T(lg_chunk)
+		OPT_WRITE_CHAR_P(dss)
 		OPT_WRITE_SIZE_T(narenas)
 		OPT_WRITE_SSIZE_T(lg_dirty_mult)
 		OPT_WRITE_BOOL(stats_print)
 		OPT_WRITE_BOOL(junk)
+		OPT_WRITE_SIZE_T(quarantine)
+		OPT_WRITE_BOOL(redzone)
 		OPT_WRITE_BOOL(zero)
-		OPT_WRITE_BOOL(sysv)
+		OPT_WRITE_BOOL(utrace)
+		OPT_WRITE_BOOL(valgrind)
 		OPT_WRITE_BOOL(xmalloc)
 		OPT_WRITE_BOOL(tcache)
-		OPT_WRITE_SSIZE_T(lg_tcache_gc_sweep)
 		OPT_WRITE_SSIZE_T(lg_tcache_max)
 		OPT_WRITE_BOOL(prof)
 		OPT_WRITE_CHAR_P(prof_prefix)
-		OPT_WRITE_SIZE_T(lg_prof_bt_max)
 		OPT_WRITE_BOOL(prof_active)
 		OPT_WRITE_SSIZE_T(lg_prof_sample)
 		OPT_WRITE_BOOL(prof_accum)
-		OPT_WRITE_SSIZE_T(lg_prof_tcmax)
 		OPT_WRITE_SSIZE_T(lg_prof_interval)
 		OPT_WRITE_BOOL(prof_gdump)
+		OPT_WRITE_BOOL(prof_final)
 		OPT_WRITE_BOOL(prof_leak)
-		OPT_WRITE_BOOL(overcommit)
 
 #undef OPT_WRITE_BOOL
 #undef OPT_WRITE_SIZE_T
 #undef OPT_WRITE_SSIZE_T
 #undef OPT_WRITE_CHAR_P
 
-		write_cb(cbopaque, "CPUs: ");
-		write_cb(cbopaque, u2s(ncpus, 10, s));
-		write_cb(cbopaque, "\n");
+		malloc_cprintf(write_cb, cbopaque, "CPUs: %u\n", ncpus);
 
 		CTL_GET("arenas.narenas", &uv, unsigned);
-		write_cb(cbopaque, "Max arenas: ");
-		write_cb(cbopaque, u2s(uv, 10, s));
-		write_cb(cbopaque, "\n");
+		malloc_cprintf(write_cb, cbopaque, "Arenas: %u\n", uv);
 
-		write_cb(cbopaque, "Pointer size: ");
-		write_cb(cbopaque, u2s(sizeof(void *), 10, s));
-		write_cb(cbopaque, "\n");
+		malloc_cprintf(write_cb, cbopaque, "Pointer size: %zu\n",
+		    sizeof(void *));
 
 		CTL_GET("arenas.quantum", &sv, size_t);
-		write_cb(cbopaque, "Quantum size: ");
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "\n");
-
-		CTL_GET("arenas.cacheline", &sv, size_t);
-		write_cb(cbopaque, "Cacheline size (assumed): ");
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "\n");
-
-		CTL_GET("arenas.subpage", &sv, size_t);
-		write_cb(cbopaque, "Subpage spacing: ");
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "\n");
-
-		if ((err = JEMALLOC_P(mallctl)("arenas.tspace_min", &sv, &ssz,
-		    NULL, 0)) == 0) {
-			write_cb(cbopaque, "Tiny 2^n-spaced sizes: [");
-			write_cb(cbopaque, u2s(sv, 10, s));
-			write_cb(cbopaque, "..");
-
-			CTL_GET("arenas.tspace_max", &sv, size_t);
-			write_cb(cbopaque, u2s(sv, 10, s));
-			write_cb(cbopaque, "]\n");
-		}
+		malloc_cprintf(write_cb, cbopaque, "Quantum size: %zu\n", sv);
 
-		CTL_GET("arenas.qspace_min", &sv, size_t);
-		write_cb(cbopaque, "Quantum-spaced sizes: [");
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "..");
-		CTL_GET("arenas.qspace_max", &sv, size_t);
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "]\n");
-
-		CTL_GET("arenas.cspace_min", &sv, size_t);
-		write_cb(cbopaque, "Cacheline-spaced sizes: [");
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "..");
-		CTL_GET("arenas.cspace_max", &sv, size_t);
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "]\n");
-
-		CTL_GET("arenas.sspace_min", &sv, size_t);
-		write_cb(cbopaque, "Subpage-spaced sizes: [");
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "..");
-		CTL_GET("arenas.sspace_max", &sv, size_t);
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, "]\n");
+		CTL_GET("arenas.page", &sv, size_t);
+		malloc_cprintf(write_cb, cbopaque, "Page size: %zu\n", sv);
 
 		CTL_GET("opt.lg_dirty_mult", &ssv, ssize_t);
 		if (ssv >= 0) {
-			write_cb(cbopaque,
-			    "Min active:dirty page ratio per arena: ");
-			write_cb(cbopaque, u2s((1U << ssv), 10, s));
-			write_cb(cbopaque, ":1\n");
+			malloc_cprintf(write_cb, cbopaque,
+			    "Min active:dirty page ratio per arena: %u:1\n",
+			    (1U << ssv));
 		} else {
-			write_cb(cbopaque,
+			malloc_cprintf(write_cb, cbopaque,
 			    "Min active:dirty page ratio per arena: N/A\n");
 		}
-		if ((err = JEMALLOC_P(mallctl)("arenas.tcache_max", &sv,
-		    &ssz, NULL, 0)) == 0) {
-			write_cb(cbopaque,
-			    "Maximum thread-cached size class: ");
-			write_cb(cbopaque, u2s(sv, 10, s));
-			write_cb(cbopaque, "\n");
-		}
-		if ((err = JEMALLOC_P(mallctl)("opt.lg_tcache_gc_sweep", &ssv,
-		    &ssz, NULL, 0)) == 0) {
-			size_t tcache_gc_sweep = (1U << ssv);
-			bool tcache_enabled;
-			CTL_GET("opt.tcache", &tcache_enabled, bool);
-			write_cb(cbopaque, "Thread cache GC sweep interval: ");
-			write_cb(cbopaque, tcache_enabled && ssv >= 0 ?
-			    u2s(tcache_gc_sweep, 10, s) : "N/A");
-			write_cb(cbopaque, "\n");
+		if ((err = je_mallctl("arenas.tcache_max", &sv, &ssz, NULL, 0))
+		    == 0) {
+			malloc_cprintf(write_cb, cbopaque,
+			    "Maximum thread-cached size class: %zu\n", sv);
 		}
-		if ((err = JEMALLOC_P(mallctl)("opt.prof", &bv, &bsz, NULL, 0))
-		   == 0 && bv) {
-			CTL_GET("opt.lg_prof_bt_max", &sv, size_t);
-			write_cb(cbopaque, "Maximum profile backtrace depth: ");
-			write_cb(cbopaque, u2s((1U << sv), 10, s));
-			write_cb(cbopaque, "\n");
-
-			CTL_GET("opt.lg_prof_tcmax", &ssv, ssize_t);
-			write_cb(cbopaque,
-			    "Maximum per thread backtrace cache: ");
-			if (ssv >= 0) {
-				write_cb(cbopaque, u2s((1U << ssv), 10, s));
-				write_cb(cbopaque, " (2^");
-				write_cb(cbopaque, u2s(ssv, 10, s));
-				write_cb(cbopaque, ")\n");
-			} else
-				write_cb(cbopaque, "N/A\n");
-
+		if ((err = je_mallctl("opt.prof", &bv, &bsz, NULL, 0)) == 0 &&
+		    bv) {
 			CTL_GET("opt.lg_prof_sample", &sv, size_t);
-			write_cb(cbopaque, "Average profile sample interval: ");
-			write_cb(cbopaque, u2s((((uint64_t)1U) << sv), 10, s));
-			write_cb(cbopaque, " (2^");
-			write_cb(cbopaque, u2s(sv, 10, s));
-			write_cb(cbopaque, ")\n");
+			malloc_cprintf(write_cb, cbopaque,
+			    "Average profile sample interval: %"PRIu64
+			    " (2^%zu)\n", (((uint64_t)1U) << sv), sv);
 
 			CTL_GET("opt.lg_prof_interval", &ssv, ssize_t);
-			write_cb(cbopaque, "Average profile dump interval: ");
 			if (ssv >= 0) {
-				write_cb(cbopaque, u2s((((uint64_t)1U) << ssv),
-				    10, s));
-				write_cb(cbopaque, " (2^");
-				write_cb(cbopaque, u2s(ssv, 10, s));
-				write_cb(cbopaque, ")\n");
-			} else
-				write_cb(cbopaque, "N/A\n");
+				malloc_cprintf(write_cb, cbopaque,
+				    "Average profile dump interval: %"PRIu64
+				    " (2^%zd)\n",
+				    (((uint64_t)1U) << ssv), ssv);
+			} else {
+				malloc_cprintf(write_cb, cbopaque,
+				    "Average profile dump interval: N/A\n");
+			}
 		}
-		CTL_GET("arenas.chunksize", &sv, size_t);
-		write_cb(cbopaque, "Chunk size: ");
-		write_cb(cbopaque, u2s(sv, 10, s));
 		CTL_GET("opt.lg_chunk", &sv, size_t);
-		write_cb(cbopaque, " (2^");
-		write_cb(cbopaque, u2s(sv, 10, s));
-		write_cb(cbopaque, ")\n");
+		malloc_cprintf(write_cb, cbopaque, "Chunk size: %zu (2^%zu)\n",
+		    (ZU(1) << sv), sv);
 	}
 
-#ifdef JEMALLOC_STATS
-	{
-		int err;
-		size_t ssz;
+	if (config_stats) {
+		size_t *cactive;
 		size_t allocated, active, mapped;
-		size_t chunks_current, chunks_high, swap_avail;
+		size_t chunks_current, chunks_high;
 		uint64_t chunks_total;
 		size_t huge_allocated;
 		uint64_t huge_nmalloc, huge_ndalloc;
 
-		ssz = sizeof(size_t);
-
+		CTL_GET("stats.cactive", &cactive, size_t *);
 		CTL_GET("stats.allocated", &allocated, size_t);
 		CTL_GET("stats.active", &active, size_t);
 		CTL_GET("stats.mapped", &mapped, size_t);
 		malloc_cprintf(write_cb, cbopaque,
-		    "Allocated: %zu, active: %zu, mapped: %zu\n", allocated,
-		    active, mapped);
+		    "Allocated: %zu, active: %zu, mapped: %zu\n",
+		    allocated, active, mapped);
+		malloc_cprintf(write_cb, cbopaque,
+		    "Current active ceiling: %zu\n", atomic_read_z(cactive));
 
 		/* Print chunk stats. */
 		CTL_GET("stats.chunks.total", &chunks_total, uint64_t);
 		CTL_GET("stats.chunks.high", &chunks_high, size_t);
 		CTL_GET("stats.chunks.current", &chunks_current, size_t);
-		if ((err = JEMALLOC_P(mallctl)("swap.avail", &swap_avail, &ssz,
-		    NULL, 0)) == 0) {
-			size_t lg_chunk;
-
-			malloc_cprintf(write_cb, cbopaque, "chunks: nchunks   "
-			    "highchunks    curchunks   swap_avail\n");
-			CTL_GET("opt.lg_chunk", &lg_chunk, size_t);
-			malloc_cprintf(write_cb, cbopaque,
-			    "  %13"PRIu64"%13zu%13zu%13zu\n",
-			    chunks_total, chunks_high, chunks_current,
-			    swap_avail << lg_chunk);
-		} else {
-			malloc_cprintf(write_cb, cbopaque, "chunks: nchunks   "
-			    "highchunks    curchunks\n");
-			malloc_cprintf(write_cb, cbopaque,
-			    "  %13"PRIu64"%13zu%13zu\n",
-			    chunks_total, chunks_high, chunks_current);
-		}
+		malloc_cprintf(write_cb, cbopaque, "chunks: nchunks   "
+		    "highchunks    curchunks\n");
+		malloc_cprintf(write_cb, cbopaque,
+		    "  %13"PRIu64" %12zu %12zu\n",
+		    chunks_total, chunks_high, chunks_current);
 
 		/* Print huge stats. */
 		CTL_GET("stats.huge.nmalloc", &huge_nmalloc, uint64_t);
@@ -723,11 +496,11 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 
 			CTL_GET("arenas.narenas", &narenas, unsigned);
 			{
-				bool initialized[narenas];
+				VARIABLE_ARRAY(bool, initialized, narenas);
 				size_t isz;
 				unsigned i, ninitialized;
 
-				isz = sizeof(initialized);
+				isz = sizeof(bool) * narenas;
 				xmallctl("arenas.initialized", initialized,
 				    &isz, NULL, 0);
 				for (i = ninitialized = 0; i < narenas; i++) {
@@ -735,12 +508,12 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 						ninitialized++;
 				}
 
-				if (ninitialized > 1) {
+				if (ninitialized > 1 || unmerged == false) {
 					/* Print merged arena stats. */
 					malloc_cprintf(write_cb, cbopaque,
 					    "\nMerged arenas stats:\n");
 					stats_arena_print(write_cb, cbopaque,
-					    narenas);
+					    narenas, bins, large);
 				}
 			}
 		}
@@ -752,11 +525,11 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 
 			CTL_GET("arenas.narenas", &narenas, unsigned);
 			{
-				bool initialized[narenas];
+				VARIABLE_ARRAY(bool, initialized, narenas);
 				size_t isz;
 				unsigned i;
 
-				isz = sizeof(initialized);
+				isz = sizeof(bool) * narenas;
 				xmallctl("arenas.initialized", initialized,
 				    &isz, NULL, 0);
 
@@ -766,12 +539,11 @@ stats_print(void (*write_cb)(void *, const char *), void *cbopaque,
 						    cbopaque,
 						    "\narenas[%u]:\n", i);
 						stats_arena_print(write_cb,
-						    cbopaque, i);
+						    cbopaque, i, bins, large);
 					}
 				}
 			}
 		}
 	}
-#endif /* #ifdef JEMALLOC_STATS */
-	write_cb(cbopaque, "--- End jemalloc statistics ---\n");
+	malloc_cprintf(write_cb, cbopaque, "--- End jemalloc statistics ---\n");
 }