Clean up threading code some more

libc/thread/mktls.c

@@ -35,7 +35,7 @@ void Bzero(void *, size_t) asm("bzero");  // gcc bug
  * Allocates thread-local storage memory for new thread.
  * @return buffer that must be released with free()
  */
-char *_mktls(char **out_tib) {
+char *_mktls(struct CosmoTib **out_tib) {
   char *tls;
   struct CosmoTib *tib;
 
@@ -60,7 +60,7 @@ char *_mktls(char **out_tib) {
   tib->tib_tid = -1;
 
   if (out_tib) {
-    *out_tib = (char *)tib;
+    *out_tib = tib;
   }
   return tls;
 }

libc/thread/posixthread.internal.h

@@ -3,6 +3,7 @@
 #include "libc/calls/struct/sched_param.h"
 #include "libc/runtime/runtime.h"
 #include "libc/thread/thread.h"
+#include "libc/thread/tls.h"
 #if !(__ASSEMBLER__ + __LINKER__ + 0)
 COSMOPOLITAN_C_START_
 
@@ -61,8 +62,8 @@ struct PosixThread {
   int tid;
   int *ctid;
   char *tls;
-  char *tib;
   char *altstack;
+  struct CosmoTib *tib;
   _Atomic(enum PosixThreadStatus) status;
   jmp_buf exiter;
   pthread_attr_t attr;

libc/thread/pthread_create.c

@@ -53,7 +53,7 @@ STATIC_YOINK("nsync_mu_unlock");
 #define MAP_STACK_OPENBSD 0x4000
 
 void _pthread_wait(struct PosixThread *pt) {
-  _wait0(pt->ctid);
+  _wait0(&pt->tib->tib_tid);
 }
 
 void _pthread_free(struct PosixThread *pt) {
@@ -206,9 +206,6 @@ int pthread_create(pthread_t *thread, const pthread_attr_t *attr,
     return EAGAIN;
   }
 
-  // child thread id is also a condition variable
-  pt->ctid = (int *)(pt->tib + 0x38);
-
   // setup attributes
   if (attr) {
     pt->attr = *attr;
@@ -301,7 +298,7 @@ int pthread_create(pthread_t *thread, const pthread_attr_t *attr,
             CLONE_VM | CLONE_THREAD | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
                 CLONE_SETTLS | CLONE_PARENT_SETTID | CLONE_CHILD_SETTID |
                 CLONE_CHILD_CLEARTID,
-            pt, &pt->tid, pt->tib, pt->ctid) == -1) {
+            pt, &pt->tid, pt->tib, &pt->tib->tib_tid) == -1) {
     rc = errno;
     _pthread_free(pt);
     errno = e;

libc/thread/spawn.c

@@ -111,7 +111,6 @@ int _spawn(int fun(void *, int), void *arg, struct spawn *opt_out_thread) {
   if (!(th->tls = _mktls(&th->tib))) {
     return -1;
   }
-  th->ctid = (int *)(th->tib + 0x38);
 
   // we must use _mapstack() to allocate the stack because OpenBSD has
   // very strict requirements for what's allowed to be used for stacks
@@ -127,7 +126,7 @@ int _spawn(int fun(void *, int), void *arg, struct spawn *opt_out_thread) {
             CLONE_VM | CLONE_THREAD | CLONE_FS | CLONE_FILES | CLONE_SIGHAND |
                 CLONE_SETTLS | CLONE_PARENT_SETTID | CLONE_CHILD_SETTID |
                 CLONE_CHILD_CLEARTID,
-            spawner, &th->ptid, th->tib, th->ctid) == -1) {
+            spawner, &th->ptid, th->tib, &th->tib->tib_tid) == -1) {
     _freestack(th->stk);
     free(th->tls);
     return -1;
@@ -143,9 +142,9 @@ int _spawn(int fun(void *, int), void *arg, struct spawn *opt_out_thread) {
  */
 int _join(struct spawn *th) {
   int rc;
-  if (th->ctid) {
+  if (th->tib) {
     // wait for ctid to become zero
-    _wait0(th->ctid);
+    _wait0(&th->tib->tib_tid);
     // free thread memory
     free(th->tls);
     rc = munmap(th->stk, GetStackSize());

libc/thread/spawn.h

@@ -1,19 +1,19 @@
 #ifndef COSMOPOLITAN_LIBC_THREAD_SPAWN_H_
 #define COSMOPOLITAN_LIBC_THREAD_SPAWN_H_
+#include "libc/thread/tls.h"
 #if !(__ASSEMBLER__ + __LINKER__ + 0)
 COSMOPOLITAN_C_START_
 
 struct spawn {
   int ptid;
-  int *ctid;
   char *stk;
   char *tls;
-  char *tib;
+  struct CosmoTib *tib;
 };
 
 int _spawn(int (*)(void *, int), void *, struct spawn *) hidden;
 int _join(struct spawn *) hidden;
-char *_mktls(char **) hidden;
+char *_mktls(struct CosmoTib **) hidden;
 
 COSMOPOLITAN_C_END_
 #endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */

libc/thread/thread.h

@@ -1,7 +1,7 @@
 #ifndef COSMOPOLITAN_LIBC_THREAD_THREAD_H_
 #define COSMOPOLITAN_LIBC_THREAD_THREAD_H_
 
-#define PTHREAD_KEYS_MAX              64
+#define PTHREAD_KEYS_MAX              128
 #define PTHREAD_STACK_MIN             FRAMESIZE
 #define PTHREAD_DESTRUCTOR_ITERATIONS 4
 
@@ -51,8 +51,10 @@ typedef struct pthread_spinlock_s {
 
 typedef struct pthread_mutex_s {
   _Atomic(int32_t) _lock;
-  uint16_t _type;
-  uint16_t _pshared;
+  unsigned _type : 2;
+  unsigned _pshared : 1;
+  unsigned _depth : 8;
+  unsigned _owner : 21;
   void *_nsync;
 } pthread_mutex_t;
 

libc/thread/tls.h

@@ -1,5 +1,6 @@
 #ifndef COSMOPOLITAN_LIBC_THREAD_TLS_H_
 #define COSMOPOLITAN_LIBC_THREAD_TLS_H_
+#include "libc/thread/thread.h"
 
 #define TLS_ALIGNMENT 64
 
@@ -22,6 +23,15 @@ struct CosmoTib {
   struct CosmoTib *tib_self2;    /* 0x30 */
   _Atomic(int32_t) tib_tid;      /* 0x38 */
   int32_t tib_errno;             /* 0x3c */
+  void *tib_nsync;
+  void *tib_reserved1;
+  void *tib_reserved2;
+  void *tib_reserved3;
+  void *tib_reserved4;
+  void *tib_reserved5;
+  void *tib_reserved6;
+  void *tib_reserved7;
+  void *tib_keys[PTHREAD_KEYS_MAX];
 };
 
 extern int __threaded;

libc/thread/wait0.c

@@ -16,17 +16,18 @@
 │ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
-#include "libc/intrin/strace.internal.h"
+#include "libc/atomic.h"
 #include "libc/calls/struct/timespec.h"
 #include "libc/dce.h"
 #include "libc/errno.h"
 #include "libc/intrin/atomic.h"
 #include "libc/intrin/describeflags.internal.h"
+#include "libc/intrin/strace.internal.h"
 #include "libc/sysv/consts/futex.h"
 #include "libc/thread/thread.h"
 #include "libc/thread/wait0.internal.h"
 
-int _futex(int *, int, int, const struct timespec *);
+int _futex(atomic_int *, int, int, const struct timespec *);
 
 /**
  * Blocks until memory location becomes zero.
@@ -35,7 +36,7 @@ int _futex(int *, int, int, const struct timespec *);
  * by the _spawn() system call when a thread terminates. The purpose of
  * this operation is to know when it's safe to munmap() a threads stack
  */
-void _wait0(const int *ctid) {
+void _wait0(const atomic_int *ctid) {
   int x, rc;
   char buf[12];
   for (;;) {

libc/thread/wait0.internal.h

@@ -1,9 +1,10 @@
 #ifndef COSMOPOLITAN_LIBC_INTRIN_WAIT0_H_
 #define COSMOPOLITAN_LIBC_INTRIN_WAIT0_H_
+#include "libc/atomic.h"
 #if !(__ASSEMBLER__ + __LINKER__ + 0)
 COSMOPOLITAN_C_START_
 
-void _wait0(const int *) hidden;
+void _wait0(const atomic_int *) hidden;
 
 COSMOPOLITAN_C_END_
 #endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */