Use dynamic memory for *NSYNC waiters

libc/errno.h

@@ -28,11 +28,11 @@ COSMOPOLITAN_C_START_
 /* this header is included by 700+ files; therefore we */
 /* hand-roll &__get_tls()->tib_errno to avoid #include */
 /* cosmopolitan uses x28 as the tls register b/c apple */
-#define errno                            \
-  (*({                                   \
-    errno_t *__ep;                       \
-    asm("sub\t%0,x28,#68" : "=r"(__ep)); \
-    __ep;                                \
+#define errno                                  \
+  (*({                                         \
+    errno_t *__ep;                             \
+    asm("sub\t%0,x28,#192-0x3c" : "=r"(__ep)); \
+    __ep;                                      \
   }))
 #else
 #define errno (*__errno_location())

libc/nexgen32e/gc.S

@@ -67,7 +67,7 @@ __gc:	.ftrace2
 
 //	if this code fails
 //	check if CosmoTib's size changed
-	sub	x8,x28,#128			// __get_tls()
+	sub	x8,x28,#192			// __get_tls()
 	ldr	x9,[x8,0x18]			// tib::garbages
 	ldr	x10,[x9]			// g->i
 	ldr	x8,[x9,8]			// g->p

libc/proc/vfork.S

@@ -112,7 +112,7 @@ vfork:
 //	} else {
 //	  __get_tls()->tib_flags &= ~TIB_FLAG_VFORKED;
 //	}
-	sub	x1,x28,#0x80		// RELIES ON TLS TIB ABI!
+	sub	x1,x28,#192		// sizeof(CosmoTib)
 	ldr	x2,[x1,64]
 	cbnz	x0,2f
 	orr	x2,x2,#TIB_FLAG_VFORKED

libc/thread/pthread_exit.c

@@ -32,6 +32,7 @@
 #include "libc/thread/thread.h"
 #include "libc/thread/tls.h"
 #include "third_party/nsync/futex.internal.h"
+#include "third_party/nsync/wait_s.internal.h"
 
 void _pthread_unwind(struct PosixThread *pt) {
   struct _pthread_cleanup_buffer *cb;
@@ -115,6 +116,9 @@ wontreturn void pthread_exit(void *rc) {
     _weaken(__cxa_thread_finalize)();
   }
   _pthread_unkey(tib);
+  if (tib->tib_nsync) {
+    nsync_waiter_destroy(tib->tib_nsync);
+  }
   _pthread_ungarbage();
   _pthread_decimate();
 

libc/thread/tls.h

@@ -37,6 +37,8 @@ struct CosmoTib {
   uint32_t tib_sigstack_size;
   uint32_t tib_sigstack_flags;
   void **tib_keys;
+  void *tib_nsync;
+  void *tib_todo[63];
 };
 
 extern int __threaded;