diff --git a/arch/arm/atomic.h b/arch/arm/atomic.h
index 302e6d8..738f528 100644
--- a/arch/arm/atomic.h
+++ b/arch/arm/atomic.h
@@ -103,6 +103,7 @@ static inline void a_store(volatile int *p, int x)
 
 static inline void a_spin()
 {
+	__k_cas(0, 0, &(int){0});
 }
 
 static inline void a_crash()
diff --git a/arch/arm/bits/alltypes.h.in b/arch/arm/bits/alltypes.h.in
index 0d750cc..183c4c4 100644
--- a/arch/arm/bits/alltypes.h.in
+++ b/arch/arm/bits/alltypes.h.in
@@ -13,6 +13,8 @@ TYPEDEF unsigned wint_t;
 TYPEDEF float float_t;
 TYPEDEF double double_t;
 
+TYPEDEF struct { long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long time_t;
 TYPEDEF long suseconds_t;
 
diff --git a/arch/i386/bits/alltypes.h.in b/arch/i386/bits/alltypes.h.in
index 502c882..8ba8f6f 100644
--- a/arch/i386/bits/alltypes.h.in
+++ b/arch/i386/bits/alltypes.h.in
@@ -27,6 +27,8 @@ TYPEDEF long double float_t;
 TYPEDEF long double double_t;
 #endif
 
+TYPEDEF struct { _Alignas(8) long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long time_t;
 TYPEDEF long suseconds_t;
 
diff --git a/arch/microblaze/atomic.h b/arch/microblaze/atomic.h
index 96265fe..abb79b5 100644
--- a/arch/microblaze/atomic.h
+++ b/arch/microblaze/atomic.h
@@ -97,6 +97,7 @@ static inline void a_store(volatile int *p, int x)
 
 static inline void a_spin()
 {
+	a_cas(&(int){0}, 0, 0);
 }
 
 static inline void a_crash()
diff --git a/arch/microblaze/bits/alltypes.h.in b/arch/microblaze/bits/alltypes.h.in
index 4657d14..a03e1b8 100644
--- a/arch/microblaze/bits/alltypes.h.in
+++ b/arch/microblaze/bits/alltypes.h.in
@@ -13,6 +13,8 @@ TYPEDEF unsigned wint_t;
 TYPEDEF float float_t;
 TYPEDEF double double_t;
 
+TYPEDEF struct { long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long time_t;
 TYPEDEF long suseconds_t;
 
diff --git a/arch/mips/atomic.h b/arch/mips/atomic.h
index 3ec0358..cc5bf49 100644
--- a/arch/mips/atomic.h
+++ b/arch/mips/atomic.h
@@ -137,6 +137,7 @@ static inline void a_store(volatile int *p, int x)
 
 static inline void a_spin()
 {
+	a_cas(&(int){0}, 0, 0);
 }
 
 static inline void a_crash()
diff --git a/arch/mips/bits/alltypes.h.in b/arch/mips/bits/alltypes.h.in
index 4657d14..a03e1b8 100644
--- a/arch/mips/bits/alltypes.h.in
+++ b/arch/mips/bits/alltypes.h.in
@@ -13,6 +13,8 @@ TYPEDEF unsigned wint_t;
 TYPEDEF float float_t;
 TYPEDEF double double_t;
 
+TYPEDEF struct { long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long time_t;
 TYPEDEF long suseconds_t;
 
diff --git a/arch/or1k/atomic.h b/arch/or1k/atomic.h
index 5b0411b..f9e6981 100644
--- a/arch/or1k/atomic.h
+++ b/arch/or1k/atomic.h
@@ -74,6 +74,7 @@ static inline void a_store(volatile int *p, int x)
 
 static inline void a_spin()
 {
+	a_cas(&(int){0}, 0, 0);
 }
 
 static inline void a_crash()
diff --git a/arch/or1k/bits/alltypes.h.in b/arch/or1k/bits/alltypes.h.in
index 0d750cc..183c4c4 100644
--- a/arch/or1k/bits/alltypes.h.in
+++ b/arch/or1k/bits/alltypes.h.in
@@ -13,6 +13,8 @@ TYPEDEF unsigned wint_t;
 TYPEDEF float float_t;
 TYPEDEF double double_t;
 
+TYPEDEF struct { long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long time_t;
 TYPEDEF long suseconds_t;
 
diff --git a/arch/powerpc/atomic.h b/arch/powerpc/atomic.h
index 1044886..1c50361 100644
--- a/arch/powerpc/atomic.h
+++ b/arch/powerpc/atomic.h
@@ -80,6 +80,7 @@ static inline void a_store(volatile int *p, int x)
 
 static inline void a_spin()
 {
+	a_cas(&(int){0}, 0, 0);
 }
 
 static inline void a_crash()
diff --git a/arch/powerpc/bits/alltypes.h.in b/arch/powerpc/bits/alltypes.h.in
index 378124c..ee7f137 100644
--- a/arch/powerpc/bits/alltypes.h.in
+++ b/arch/powerpc/bits/alltypes.h.in
@@ -13,6 +13,8 @@ TYPEDEF unsigned wint_t;
 TYPEDEF float float_t;
 TYPEDEF double double_t;
 
+TYPEDEF struct { long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long time_t;
 TYPEDEF long suseconds_t;
 
diff --git a/arch/sh/atomic.h b/arch/sh/atomic.h
index 93ab54f..b95bbff 100644
--- a/arch/sh/atomic.h
+++ b/arch/sh/atomic.h
@@ -53,6 +53,7 @@ static inline void a_dec(volatile int *x)
 
 static inline void a_spin()
 {
+	a_cas(&(int){0}, 0, 0);
 }
 
 static inline void a_crash()
diff --git a/arch/sh/bits/alltypes.h.in b/arch/sh/bits/alltypes.h.in
index 378124c..ee7f137 100644
--- a/arch/sh/bits/alltypes.h.in
+++ b/arch/sh/bits/alltypes.h.in
@@ -13,6 +13,8 @@ TYPEDEF unsigned wint_t;
 TYPEDEF float float_t;
 TYPEDEF double double_t;
 
+TYPEDEF struct { long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long time_t;
 TYPEDEF long suseconds_t;
 
diff --git a/arch/x32/bits/alltypes.h.in b/arch/x32/bits/alltypes.h.in
index 8930efa..8e396c9 100644
--- a/arch/x32/bits/alltypes.h.in
+++ b/arch/x32/bits/alltypes.h.in
@@ -18,6 +18,8 @@ TYPEDEF float float_t;
 TYPEDEF double double_t;
 #endif
 
+TYPEDEF struct { long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long long time_t;
 TYPEDEF long long suseconds_t;
 
diff --git a/arch/x86_64/bits/alltypes.h.in b/arch/x86_64/bits/alltypes.h.in
index 34b7d6a..7b4f3e7 100644
--- a/arch/x86_64/bits/alltypes.h.in
+++ b/arch/x86_64/bits/alltypes.h.in
@@ -18,6 +18,8 @@ TYPEDEF float float_t;
 TYPEDEF double double_t;
 #endif
 
+TYPEDEF struct { long long __ll; long double __ld; } max_align_t;
+
 TYPEDEF long time_t;
 TYPEDEF long suseconds_t;
 
diff --git a/include/assert.h b/include/assert.h
index ab745db..1ee02a4 100644
--- a/include/assert.h
+++ b/include/assert.h
@@ -8,6 +8,10 @@
 #define assert(x) ((void)((x) || (__assert_fail(#x, __FILE__, __LINE__, __func__),0)))
 #endif
 
+#ifndef __cplusplus
+#define static_assert _Static_assert
+#endif
+
 #ifdef __cplusplus
 extern "C" {
 #endif
diff --git a/include/float.h b/include/float.h
index 2b2ad39..161e167 100644
--- a/include/float.h
+++ b/include/float.h
@@ -11,8 +11,10 @@
 #define FLT_MANT_DIG 24
 #define FLT_MIN_EXP (-125)
 #define FLT_MAX_EXP 128
+#define FLT_HAS_SUBNORM 1
 
 #define FLT_DIG 6
+#define FLT_DECIMAL_DIG 9
 #define FLT_MIN_10_EXP (-37)
 #define FLT_MAX_10_EXP 38
 
@@ -24,11 +26,16 @@
 #define DBL_MANT_DIG 53
 #define DBL_MIN_EXP (-1021)
 #define DBL_MAX_EXP 1024
+#define DBL_HAS_SUBNORM 1
 
 #define DBL_DIG 15
+#define DBL_DECIMAL_DIG 17
 #define DBL_MIN_10_EXP (-307)
 #define DBL_MAX_10_EXP 308
 
+#define LDBL_HAS_SUBNORM 1
+#define LDBL_DECIMAL_DIG DECIMAL_DIG
+
 #include <bits/float.h>
 
 #endif
diff --git a/include/malloc.h b/include/malloc.h
index e69de29..35f8b19 100644
--- a/include/malloc.h
+++ b/include/malloc.h
@@ -0,0 +1,25 @@
+#ifndef _MALLOC_H
+#define _MALLOC_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#define __NEED_size_t
+
+#include <bits/alltypes.h>
+
+void *malloc (size_t);
+void *calloc (size_t, size_t);
+void *realloc (void *, size_t);
+void free (void *);
+void *valloc (size_t);
+void *memalign(size_t, size_t);
+
+size_t malloc_usable_size(void *);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/include/stdalign.h b/include/stdalign.h
index b6e50ae..2cc94be 100644
--- a/include/stdalign.h
+++ b/include/stdalign.h
@@ -1,6 +1,8 @@
 #ifndef _STDALIGN_H
 #define _STDALIGN_H
 
+#ifndef __cplusplus
+
 /* this whole header only works in C11 or with compiler extensions */
 #if __STDC_VERSION__ < 201112L && defined( __GNUC__)
 #define _Alignas(t) __attribute__((__aligned__(t)))
@@ -9,6 +11,9 @@
 
 #define alignas _Alignas
 #define alignof _Alignof
+
+#endif
+
 #define __alignas_is_defined 1
 #define __alignof_is_defined 1
 
diff --git a/include/stddef.h b/include/stddef.h
index 0a32919..bd75385 100644
--- a/include/stddef.h
+++ b/include/stddef.h
@@ -10,6 +10,9 @@
 #define __NEED_ptrdiff_t
 #define __NEED_size_t
 #define __NEED_wchar_t
+#if __STDC_VERSION__ >= 201112L || __cplusplus >= 201103L
+#define __NEED_max_align_t
+#endif
 
 #include <bits/alltypes.h>
 
diff --git a/include/stdnoreturn.h b/include/stdnoreturn.h
index 60d924a..5c6aeeb 100644
--- a/include/stdnoreturn.h
+++ b/include/stdnoreturn.h
@@ -1,5 +1,7 @@
 #ifndef _STDNORETURN_H
 #define _STDNORETURN_H
+#ifndef __cplusplus
 #include <features.h>
 #define noreturn _Noreturn
 #endif
+#endif
diff --git a/src/internal/pthread_impl.h b/src/internal/pthread_impl.h
index 2d090f8..74c62cc 100644
--- a/src/internal/pthread_impl.h
+++ b/src/internal/pthread_impl.h
@@ -44,6 +44,7 @@ struct pthread {
 	int exitlock[2];
 	int startlock[2];
 	unsigned long sigmask[_NSIG/8/sizeof(long)];
+	void *stdio_locks;
 };
 
 struct __timer {
@@ -112,7 +113,7 @@ static inline void __wake(volatile void *addr, int cnt, int priv)
 {
 	if (priv) priv = 128;
 	if (cnt<0) cnt = INT_MAX;
-	__syscall(SYS_futex, addr, FUTEX_WAKE|priv, cnt) != -EINVAL ||
+	__syscall(SYS_futex, addr, FUTEX_WAKE|priv, cnt) != -ENOSYS ||
 	__syscall(SYS_futex, addr, FUTEX_WAKE, cnt);
 }
 
diff --git a/src/internal/stdio_impl.h b/src/internal/stdio_impl.h
index 79be9fd..d659522 100644
--- a/src/internal/stdio_impl.h
+++ b/src/internal/stdio_impl.h
@@ -46,6 +46,7 @@ struct _IO_FILE {
 	void *mustbezero_2;
 	unsigned char *shend;
 	off_t shlim, shcnt;
+	FILE *prev_locked, *next_locked;
 };
 
 size_t __stdio_read(FILE *, unsigned char *, size_t);
diff --git a/src/malloc/malloc_usable_size.c b/src/malloc/malloc_usable_size.c
new file mode 100644
index 0000000..8cccd9d
--- /dev/null
+++ b/src/malloc/malloc_usable_size.c
@@ -0,0 +1,17 @@
+#include <malloc.h>
+
+void *(*const __realloc_dep)(void *, size_t) = realloc;
+
+struct chunk {
+	size_t psize, csize;
+	struct chunk *next, *prev;
+};
+
+#define OVERHEAD (2*sizeof(size_t))
+#define CHUNK_SIZE(c) ((c)->csize & -2)
+#define MEM_TO_CHUNK(p) (struct chunk *)((char *)(p) - OVERHEAD)
+
+size_t malloc_usable_size(void *p)
+{
+	return CHUNK_SIZE(MEM_TO_CHUNK(p)) - OVERHEAD;
+}
diff --git a/src/network/dn_expand.c b/src/network/dn_expand.c
index 849df19..d9b3393 100644
--- a/src/network/dn_expand.c
+++ b/src/network/dn_expand.c
@@ -4,11 +4,13 @@
 int __dn_expand(const unsigned char *base, const unsigned char *end, const unsigned char *src, char *dest, int space)
 {
 	const unsigned char *p = src;
-	char *dend = dest + (space > 254 ? 254 : space);
+	char *dend, *dbegin = dest;
 	int len = -1, i, j;
-	if (p==end || !*p) return -1;
+	if (p==end || space <= 0) return -1;
+	dend = dest + (space > 254 ? 254 : space);
 	/* detect reference loop using an iteration counter */
 	for (i=0; i < end-base; i+=2) {
+		/* loop invariants: p<end, dest<dend */
 		if (*p & 0xc0) {
 			if (p+1==end) return -1;
 			j = ((p[0] & 0x3f) << 8) | p[1];
@@ -16,11 +18,12 @@ int __dn_expand(const unsigned char *base, const unsigned char *end, const unsig
 			if (j >= end-base) return -1;
 			p = base+j;
 		} else if (*p) {
-			j = *p+1;
-			if (j>=end-p || j>dend-dest) return -1;
-			while (--j) *dest++ = *++p;
-			*dest++ = *++p ? '.' : 0;
+			if (dest != dbegin) *dest++ = '.';
+			j = *p++;
+			if (j >= end-p || j >= dend-dest) return -1;
+			while (j--) *dest++ = *p++;
 		} else {
+			*dest = 0;
 			if (len < 0) len = p+1-src;
 			return len;
 		}
diff --git a/src/stdio/fclose.c b/src/stdio/fclose.c
index 38e8a1e..317b3c9 100644
--- a/src/stdio/fclose.c
+++ b/src/stdio/fclose.c
@@ -1,4 +1,8 @@
 #include "stdio_impl.h"
+#include "libc.h"
+
+static void dummy(FILE *f) { }
+weak_alias(dummy, __unlist_locked_file);
 
 int fclose(FILE *f)
 {
@@ -7,6 +11,8 @@ int fclose(FILE *f)
 	
 	FFINALLOCK(f);
 
+	__unlist_locked_file(f);
+
 	if (!(perm = f->flags & F_PERM)) {
 		OFLLOCK();
 		if (f->prev) f->prev->next = f->next;
diff --git a/src/stdio/fgets.c b/src/stdio/fgets.c
index b01a418..d3f9819 100644
--- a/src/stdio/fgets.c
+++ b/src/stdio/fgets.c
@@ -10,14 +10,16 @@ char *fgets(char *restrict s, int n, FILE *restrict f)
 	size_t k;
 	int c;
 
+	FLOCK(f);
+
 	if (n--<=1) {
+		f->mode |= f->mode-1;
+		FUNLOCK(f);
 		if (n) return 0;
 		*s = 0;
 		return s;
 	}
 
-	FLOCK(f);
-
 	while (n) {
 		z = memchr(f->rpos, '\n', f->rend - f->rpos);
 		k = z ? z - f->rpos + 1 : f->rend - f->rpos;
@@ -34,7 +36,7 @@ char *fgets(char *restrict s, int n, FILE *restrict f)
 		n--;
 		if ((*p++ = c) == '\n') break;
 	}
-	*p = 0;
+	if (s) *p = 0;
 
 	FUNLOCK(f);
 
diff --git a/src/stdio/fputs.c b/src/stdio/fputs.c
index 1112b19..4737f44 100644
--- a/src/stdio/fputs.c
+++ b/src/stdio/fputs.c
@@ -3,9 +3,7 @@
 
 int fputs(const char *restrict s, FILE *restrict f)
 {
-	size_t l = strlen(s);
-	if (!l) return 0;
-	return (int)fwrite(s, l, 1, f) - 1;
+	return (int)fwrite(s, strlen(s), 1, f) - 1;
 }
 
 weak_alias(fputs, fputs_unlocked);
diff --git a/src/stdio/fread.c b/src/stdio/fread.c
index c461256..33a65f5 100644
--- a/src/stdio/fread.c
+++ b/src/stdio/fread.c
@@ -8,11 +8,10 @@ size_t fread(void *restrict destv, size_t size, size_t nmemb, FILE *restrict f)
 	unsigned char *dest = destv;
 	size_t len = size*nmemb, l = len, k;
 
-	/* Never touch the file if length is zero.. */
-	if (!l) return 0;
-
 	FLOCK(f);
 
+	f->mode |= f->mode-1;
+
 	if (f->rend - f->rpos > 0) {
 		/* First exhaust the buffer. */
 		k = MIN(f->rend - f->rpos, l);
diff --git a/src/stdio/ftrylockfile.c b/src/stdio/ftrylockfile.c
index 56cccaf..6f9a4b8 100644
--- a/src/stdio/ftrylockfile.c
+++ b/src/stdio/ftrylockfile.c
@@ -2,9 +2,26 @@
 #include "pthread_impl.h"
 #include <limits.h>
 
+void __do_orphaned_stdio_locks()
+{
+	FILE *f;
+	for (f=__pthread_self()->stdio_locks; f; f=f->next_locked)
+		a_store(&f->lock, 0x40000000);
+}
+
+void __unlist_locked_file(FILE *f)
+{
+	if (f->lockcount) {
+		if (f->next_locked) f->next_locked->prev_locked = f->prev_locked;
+		if (f->prev_locked) f->prev_locked->next_locked = f->next_locked;
+		else __pthread_self()->stdio_locks = f->next_locked;
+	}
+}
+
 int ftrylockfile(FILE *f)
 {
-	int tid = __pthread_self()->tid;
+	pthread_t self = __pthread_self();
+	int tid = self->tid;
 	if (f->lock == tid) {
 		if (f->lockcount == LONG_MAX)
 			return -1;
@@ -15,5 +32,8 @@ int ftrylockfile(FILE *f)
 	if (f->lock || a_cas(&f->lock, 0, tid))
 		return -1;
 	f->lockcount = 1;
+	f->prev_locked = 0;
+	f->next_locked = self->stdio_locks;
+	self->stdio_locks = f;
 	return 0;
 }
diff --git a/src/stdio/funlockfile.c b/src/stdio/funlockfile.c
index f8a2a07..30a07ef 100644
--- a/src/stdio/funlockfile.c
+++ b/src/stdio/funlockfile.c
@@ -1,7 +1,15 @@
 #include "stdio_impl.h"
 #include "pthread_impl.h"
 
+void __unlist_locked_file(FILE *);
+
 void funlockfile(FILE *f)
 {
-	if (!--f->lockcount) __unlockfile(f);
+	if (f->lockcount == 1) {
+		__unlist_locked_file(f);
+		f->lockcount = 0;
+		__unlockfile(f);
+	} else {
+		f->lockcount--;
+	}
 }
diff --git a/src/stdio/fwrite.c b/src/stdio/fwrite.c
index d5f6542..81ec271 100644
--- a/src/stdio/fwrite.c
+++ b/src/stdio/fwrite.c
@@ -28,7 +28,6 @@ size_t __fwritex(const unsigned char *restrict s, size_t l, FILE *restrict f)
 size_t fwrite(const void *restrict src, size_t size, size_t nmemb, FILE *restrict f)
 {
 	size_t k, l = size*nmemb;
-	if (!l) return l;
 	FLOCK(f);
 	k = __fwritex(src, l, f);
 	FUNLOCK(f);
diff --git a/src/thread/__timedwait.c b/src/thread/__timedwait.c
index 39eb996..d6f1233 100644
--- a/src/thread/__timedwait.c
+++ b/src/thread/__timedwait.c
@@ -29,7 +29,7 @@ int __timedwait(volatile int *addr, int val,
 	pthread_cleanup_push(cleanup, arg);
 
 	r = -__syscall_cp(SYS_futex, addr, FUTEX_WAIT|priv, val, top);
-	if (r == EINVAL) r = -__syscall_cp(SYS_futex, addr, FUTEX_WAIT, val, top);
+	if (r == ENOSYS) r = -__syscall_cp(SYS_futex, addr, FUTEX_WAIT, val, top);
 	if (r != EINTR && r != ETIMEDOUT) r = 0;
 
 	pthread_cleanup_pop(0);
diff --git a/src/thread/__wait.c b/src/thread/__wait.c
index ec1e820..01ee598 100644
--- a/src/thread/__wait.c
+++ b/src/thread/__wait.c
@@ -2,15 +2,15 @@
 
 void __wait(volatile int *addr, volatile int *waiters, int val, int priv)
 {
-	int spins=10000;
+	int spins=100;
 	if (priv) priv = 128;
-	while (spins--) {
+	while (spins-- && (!waiters || !*waiters)) {
 		if (*addr==val) a_spin();
 		else return;
 	}
 	if (waiters) a_inc(waiters);
 	while (*addr==val) {
-		__syscall(SYS_futex, addr, FUTEX_WAIT|priv, val, 0) != -EINVAL
+		__syscall(SYS_futex, addr, FUTEX_WAIT|priv, val, 0) != -ENOSYS
 		|| __syscall(SYS_futex, addr, FUTEX_WAIT, val, 0);
 	}
 	if (waiters) a_dec(waiters);
diff --git a/src/thread/pthread_barrier_wait.c b/src/thread/pthread_barrier_wait.c
index 6b329c9..e15abb8 100644
--- a/src/thread/pthread_barrier_wait.c
+++ b/src/thread/pthread_barrier_wait.c
@@ -79,7 +79,7 @@ int pthread_barrier_wait(pthread_barrier_t *b)
 	/* First thread to enter the barrier becomes the "instance owner" */
 	if (!inst) {
 		struct instance new_inst = { 0 };
-		int spins = 10000;
+		int spins = 200;
 		b->_b_inst = inst = &new_inst;
 		a_store(&b->_b_lock, 0);
 		if (b->_b_waiters) __wake(&b->_b_lock, 1, 1);
@@ -87,7 +87,7 @@ int pthread_barrier_wait(pthread_barrier_t *b)
 			a_spin();
 		a_inc(&inst->finished);
 		while (inst->finished == 1)
-			__syscall(SYS_futex,&inst->finished,FUTEX_WAIT|128,1,0) != -EINTR
+			__syscall(SYS_futex,&inst->finished,FUTEX_WAIT|128,1,0) != -ENOSYS
 			|| __syscall(SYS_futex,&inst->finished,FUTEX_WAIT,1,0);
 		return PTHREAD_BARRIER_SERIAL_THREAD;
 	}
diff --git a/src/thread/pthread_cond_timedwait.c b/src/thread/pthread_cond_timedwait.c
index c5cf66c..2d192b0 100644
--- a/src/thread/pthread_cond_timedwait.c
+++ b/src/thread/pthread_cond_timedwait.c
@@ -52,7 +52,7 @@ static inline void unlock_requeue(volatile int *l, volatile int *r, int w)
 {
 	a_store(l, 0);
 	if (w) __wake(l, 1, 1);
-	else __syscall(SYS_futex, l, FUTEX_REQUEUE|128, 0, 1, r) != -EINVAL
+	else __syscall(SYS_futex, l, FUTEX_REQUEUE|128, 0, 1, r) != -ENOSYS
 		|| __syscall(SYS_futex, l, FUTEX_REQUEUE, 0, 1, r);
 }
 
diff --git a/src/thread/pthread_create.c b/src/thread/pthread_create.c
index c8c117b..e441bda 100644
--- a/src/thread/pthread_create.c
+++ b/src/thread/pthread_create.c
@@ -3,6 +3,7 @@
 #include "stdio_impl.h"
 #include "libc.h"
 #include <sys/mman.h>
+#include <string.h>
 
 static void dummy_0()
 {
@@ -11,6 +12,7 @@ weak_alias(dummy_0, __acquire_ptc);
 weak_alias(dummy_0, __release_ptc);
 weak_alias(dummy_0, __pthread_tsd_run_dtors);
 weak_alias(dummy_0, __do_private_robust_list);
+weak_alias(dummy_0, __do_orphaned_stdio_locks);
 
 _Noreturn void pthread_exit(void *result)
 {
@@ -65,6 +67,7 @@ _Noreturn void pthread_exit(void *result)
 	}
 
 	__do_private_robust_list();
+	__do_orphaned_stdio_locks();
 
 	if (self->detached && self->map_base) {
 		/* Detached threads must avoid the kernel clear_child_tid
@@ -175,6 +178,7 @@ int pthread_create(pthread_t *restrict res, const pthread_attr_t *restrict attrp
 		if (need < size/8 && need < 2048) {
 			tsd = stack - __pthread_tsd_size;
 			stack = tsd - libc.tls_size;
+			memset(stack, 0, need);
 		} else {
 			size = ROUND(need);
 			guard = 0;
diff --git a/src/thread/pthread_mutex_timedlock.c b/src/thread/pthread_mutex_timedlock.c
index 2a959d2..ae883f9 100644
--- a/src/thread/pthread_mutex_timedlock.c
+++ b/src/thread/pthread_mutex_timedlock.c
@@ -8,6 +8,12 @@ int pthread_mutex_timedlock(pthread_mutex_t *restrict m, const struct timespec *
 
 	int r, t, priv = (m->_m_type & 128) ^ 128;
 
+	r = pthread_mutex_trylock(m);
+	if (r != EBUSY) return r;
+	
+	int spins = 100;
+	while (spins-- && m->_m_lock && !m->_m_waiters) a_spin();
+
 	while ((r=pthread_mutex_trylock(m)) == EBUSY) {
 		if (!(r=m->_m_lock) || ((r&0x40000000) && (m->_m_type&4)))
 			continue;
diff --git a/src/thread/pthread_rwlock_timedrdlock.c b/src/thread/pthread_rwlock_timedrdlock.c
index a2b4d44..ea50da4 100644
--- a/src/thread/pthread_rwlock_timedrdlock.c
+++ b/src/thread/pthread_rwlock_timedrdlock.c
@@ -3,6 +3,13 @@
 int pthread_rwlock_timedrdlock(pthread_rwlock_t *restrict rw, const struct timespec *restrict at)
 {
 	int r, t;
+
+	r = pthread_rwlock_tryrdlock(rw);
+	if (r != EBUSY) return r;
+	
+	int spins = 100;
+	while (spins-- && rw->_rw_lock && !rw->_rw_waiters) a_spin();
+
 	while ((r=pthread_rwlock_tryrdlock(rw))==EBUSY) {
 		if (!(r=rw->_rw_lock) || (r&0x7fffffff)!=0x7fffffff) continue;
 		t = r | 0x80000000;
diff --git a/src/thread/pthread_rwlock_timedwrlock.c b/src/thread/pthread_rwlock_timedwrlock.c
index 63a32ec..8d04f56 100644
--- a/src/thread/pthread_rwlock_timedwrlock.c
+++ b/src/thread/pthread_rwlock_timedwrlock.c
@@ -3,6 +3,13 @@
 int pthread_rwlock_timedwrlock(pthread_rwlock_t *restrict rw, const struct timespec *restrict at)
 {
 	int r, t;
+	
+	r = pthread_rwlock_trywrlock(rw);
+	if (r != EBUSY) return r;
+	
+	int spins = 100;
+	while (spins-- && rw->_rw_lock && !rw->_rw_waiters) a_spin();
+
 	while ((r=pthread_rwlock_trywrlock(rw))==EBUSY) {
 		if (!(r=rw->_rw_lock)) continue;
 		t = r | 0x80000000;
diff --git a/src/thread/sem_timedwait.c b/src/thread/sem_timedwait.c
index bfcb6dc..b5a60ad 100644
--- a/src/thread/sem_timedwait.c
+++ b/src/thread/sem_timedwait.c
@@ -8,6 +8,11 @@ static void cleanup(void *p)
 
 int sem_timedwait(sem_t *restrict sem, const struct timespec *restrict at)
 {
+	if (!sem_trywait(sem)) return 0;
+
+	int spins = 100;
+	while (spins-- && sem->__val[0] <= 0 && !sem->__val[1]) a_spin();
+
 	while (sem_trywait(sem)) {
 		int r;
 		a_inc(sem->__val+1);