Get threads working well on MacOS Arm64

- Now using 10x better GCD semaphores
- We now generate Linux-like thread ids
- We now use fast system clock / sleep libraries
- The APE M1 loader now generates Linux-like stacks

libc/intrin/assertfail.greg.c

@@ -29,6 +29,7 @@
 #include "libc/runtime/runtime.h"
 #include "libc/runtime/symbols.internal.h"
 #include "libc/thread/thread.h"
+#include "libc/thread/tls.h"
 
 relegated void __assert_fail(const char *expr, const char *file, int line) {
   int me, owner;
@@ -37,7 +38,7 @@ relegated void __assert_fail(const char *expr, const char *file, int line) {
     strace_enabled(-1);
     ftrace_enabled(-1);
     owner = 0;
-    me = sys_gettid();
+    me = __tls_enabled ? __get_tls()->tib_tid : sys_gettid();
     pthread_setcancelstate(PTHREAD_CANCEL_DISABLE, 0);
     kprintf("%s:%d: assert(%s) failed (tid %d) %m\n", file, line, expr, me);
     if (__vforked ||

libc/intrin/exit1.greg.c

@@ -20,6 +20,7 @@
 #include "libc/intrin/asmflag.h"
 #include "libc/nt/thread.h"
 #include "libc/runtime/runtime.h"
+#include "libc/runtime/syslib.internal.h"
 #include "libc/sysv/consts/nr.h"
 #include "libc/thread/tls.h"
 
@@ -74,13 +75,16 @@ privileged wontreturn void _Exit1(int rc) {
   }
   notpossible;
 #elif defined(__aarch64__)
-  register long r0 asm("x0") = rc;
-  asm volatile("mov\tx8,%0\n\t"
-               "mov\tx16,%1\n\t"
-               "svc\t0"
-               : /* no outputs */
-               : "i"(93), "i"(0x169), "r"(r0)
-               : "x8", "memory");
+  if (IsLinux()) {
+    register long r0 asm("x0") = rc;
+    asm volatile("mov\tx8,%0\n\t"
+                 "svc\t0"
+                 : /* no outputs */
+                 : "i"(93), "r"(r0)
+                 : "x8", "memory");
+  } else if (IsXnu()) {
+    __syslib->pthread_exit(0);
+  }
   notpossible;
 #else
 #error "arch unsupported"

libc/intrin/kprintf.greg.c

@@ -227,8 +227,8 @@ privileged dontinline void klog(const char *b, size_t n) {
   register long r0 asm("x0") = (long)2;
   register long r1 asm("x1") = (long)b;
   register long r2 asm("x2") = (long)n;
-  register long r8 asm("x8") = (long)__NR_write & 0x7ff;
-  register long r16 asm("x16") = (long)__NR_write & 0x7ff;
+  register long r8 asm("x8") = (long)__NR_write;
+  register long r16 asm("x16") = (long)__NR_write;
   register long res_x0 asm("x0");
   asm volatile("svc\t0"
                : "=r"(res_x0)

libc/intrin/sched_yield.S

@@ -90,13 +90,18 @@ sched_yield:
 
 #elif defined(__aarch64__)
 
-	mov	x0,#0
-	mov	x1,#0
-	mov	x2,#0
-	mov	x3,#0
-	mov	x8,#0x7c			// sched_yield() for linux
-	mov	x16,#0x85d			// select(0,0,0,0) for xnu
+	stp	x29,x30,[sp,-32]!
+	mov	x29,sp
+	mov	x3,0
+	mov	x2,0
+	add	x4,sp,16
+	mov	x1,0
+	mov	w0,0
+	stp	xzr,xzr,[sp,16]
+	mov	x8,#0x7c			// sched_yield() for gnu/systemd
+	mov	x16,#0x5d			// select(0,0,0,0,&blah) for xnu
 	svc	0
+	ldp	x29,x30,[sp],32
 	ret
 
 #else

libc/intrin/sys_gettid.greg.c

@@ -63,14 +63,15 @@ privileged int sys_gettid(void) {
   }
   return tid;
 #elif defined(__aarch64__)
-  register long res_x0 asm("x0");
+  // this can't be used on xnu
+  if (!IsLinux()) notpossible;
+  register long res asm("x0");
   asm volatile("mov\tx8,%1\n\t"
-               "mov\tx16,%2\n\t"
                "svc\t0"
-               : "=r"(res_x0)
-               : "i"(178), "i"(186)
+               : "=r"(res)
+               : "i"(178)
                : "x8", "memory");
-  return res_x0;
+  return res;
 #else
 #error "arch unsupported"
 #endif