Refactor some code

- Write tests for cthreads
- Fix bugs in pe2.com tool
- Fix ASAN issue with GetDosEnviron()
- Consolidate the cthread header files
- Some code size optimizations for MODE=
- Attempted to squash a tls linker warning
- Attempted to get futexes working on FreeBSD

libc/thread/attr.c

@@ -18,7 +18,7 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/errno.h"
 #include "libc/runtime/stack.h"
-#include "libc/thread/attr.h"
+#include "libc/thread/thread.h"
 
 #define MIN_STACKSIZE (8 * PAGESIZE)  // includes guard, rounds up to FRAMESIZE
 #define MIN_GUARDSIZE PAGESIZE

libc/thread/attr.h

@@ -1,30 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_ATTR_H_
-#define COSMOPOLITAN_LIBC_THREAD_ATTR_H_
-
-#define CTHREAD_CREATE_DETACHED 1
-#define CTHREAD_CREATE_JOINABLE 0
-
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-/**
- * @fileoverview cosmopolitan thread attributes
- */
-
-typedef struct cthread_attr_t {
-  size_t stacksize, guardsize;
-  int mode;
-} cthread_attr_t;
-
-int cthread_attr_init(cthread_attr_t*);
-int cthread_attr_destroy(cthread_attr_t*);
-int cthread_attr_setstacksize(cthread_attr_t*, size_t);
-size_t thread_attr_getstacksize(const cthread_attr_t*);
-int cthread_attr_setguardsize(cthread_attr_t*, size_t);
-size_t cthread_attr_getguardsize(const cthread_attr_t*);
-int cthread_attr_setdetachstate(cthread_attr_t*, int);
-int cthread_attr_getdetachstate(const cthread_attr_t*);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_ATTR_H_ */

libc/thread/create.c

@@ -28,10 +28,7 @@
 #include "libc/sysv/consts/clone.h"
 #include "libc/sysv/consts/map.h"
 #include "libc/sysv/consts/prot.h"
-#include "libc/thread/attr.h"
-#include "libc/thread/create.h"
-#include "libc/thread/descriptor.h"
-#include "libc/thread/zombie.h"
+#include "libc/thread/thread.h"
 
 STATIC_YOINK("_main_thread_ctor");
 
@@ -99,8 +96,8 @@ static int cthread_start(void *arg) {
  * @return 0 on success, or error number on failure
  * @threadsafe
  */
-int cthread_create(cthread_t *restrict ptd, const cthread_attr_t *restrict attr,
-                   void *(*func)(void *), void *restrict arg) {
+int cthread_create(cthread_t *ptd, const cthread_attr_t *attr,
+                   void *(*func)(void *), void *arg) {
   int rc, tid;
   cthread_t td;
   cthread_attr_t default_attr;

libc/thread/create.h

@@ -1,17 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_CREATE_H_
-#define COSMOPOLITAN_LIBC_THREAD_CREATE_H_
-#include "libc/thread/attr.h"
-#include "libc/thread/descriptor.h"
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-/**
- * @fileoverview Create a cosmopolitan thread
- */
-
-int cthread_create(cthread_t *restrict, const cthread_attr_t *restrict,
-                   void *(*)(void *), void *restrict);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_CREATE_H_ */

libc/thread/descriptor.h

@@ -1,41 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_DESCRIPTOR_H_
-#define COSMOPOLITAN_LIBC_THREAD_DESCRIPTOR_H_
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-#include "libc/runtime/runtime.h"
-
-/**
- * @fileoverview thread types
- */
-
-enum cthread_state {
-  cthread_started = 0,
-  cthread_joining = 1,
-  cthread_finished = 2,
-  cthread_detached = 4,
-  cthread_main = 127,
-};
-
-struct cthread_descriptor_t {
-  struct cthread_descriptor_t *self;  /* 0x00 */
-  void *(*func)(void *);              /* 0x08 */
-  int32_t __pad0;                     /* 0x10 */
-  int32_t state;                      /* 0x14 */
-  void *arg;                          /* 0x18 */
-  void *pthread_ret_ptr;              /* 0x20 */
-  int64_t __pad1;                     /* 0x28 */
-  struct cthread_descriptor_t *self2; /* 0x30 */
-  int32_t tid;                        /* 0x38 */
-  int32_t err;                        /* 0x3c */
-  void *exitcode;
-  struct {
-    char *top, *bottom;
-  } stack, alloc;
-  jmp_buf exiter;
-};
-
-typedef struct cthread_descriptor_t *cthread_t;
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_DESCRIPTOR_H_ */

libc/thread/detach.c

@@ -22,8 +22,7 @@
 #include "libc/errno.h"
 #include "libc/intrin/asan.internal.h"
 #include "libc/str/str.h"
-#include "libc/thread/descriptor.h"
-#include "libc/thread/detach.h"
+#include "libc/thread/thread.h"
 
 /**
  * Detaches thread.

libc/thread/detach.h

@@ -1,16 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_DETACH_H_
-#define COSMOPOLITAN_LIBC_THREAD_DETACH_H_
-#include "libc/thread/descriptor.h"
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-/**
- * @fileoverview detach a thread
- */
-
-int cthread_detach(cthread_t);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_DETACH_H_ */
- 

libc/thread/exit.c

@@ -18,8 +18,7 @@
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/strace.internal.h"
 #include "libc/runtime/runtime.h"
-#include "libc/thread/exit.h"
-#include "libc/thread/self.h"
+#include "libc/thread/thread.h"
 
 /**
  * Exits cosmopolitan thread.

libc/thread/exit.h

@@ -1,10 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_EXIT_H_
-#define COSMOPOLITAN_LIBC_THREAD_EXIT_H_
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-wontreturn void cthread_exit(void *);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_EXIT_H_ */

libc/thread/freebsd.internal.h

@@ -3,8 +3,6 @@
 #include "libc/bits/asmflag.h"
 #include "libc/calls/struct/timespec.h"
 #include "libc/errno.h"
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
 
 /**
  * @fileoverview FreeBSD Threading
@@ -13,6 +11,13 @@ COSMOPOLITAN_C_START_
  *     maximum legal range is PID_MAX + 2 (100001) through INT_MAX
  */
 
+#define UMTX_OP_MUTEX_WAIT 17
+#define UMTX_OP_MUTEX_WAKE 18
+#define UMTX_ABSTIME       1
+
+#if !(__ASSEMBLER__ + __LINKER__ + 0)
+COSMOPOLITAN_C_START_
+
 struct rtprio {
   uint16_t type; /* scheduling class */
   uint16_t prio;
@@ -31,6 +36,14 @@ struct thr_param {
   struct rtprio *rtp;
 };
 
+struct _umtx_time {
+  struct timespec _timeout;
+  uint32_t _flags;
+  uint32_t _clockid;
+};
+
+int _umtx_op(void *, int, unsigned long, void *, void *);
+
 COSMOPOLITAN_C_END_
 #endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
 #endif /* COSMOPOLITAN_LIBC_THREAD_FREEBSD_INTERNAL_H_ */

libc/thread/init.c

@@ -27,8 +27,7 @@
 #include "libc/str/str.h"
 #include "libc/sysv/consts/map.h"
 #include "libc/sysv/consts/prot.h"
-#include "libc/thread/descriptor.h"
-#include "libc/thread/self.h"
+#include "libc/thread/thread.h"
 
 static textstartup void _main_thread_init(void) {
   _Static_assert(offsetof(struct cthread_descriptor_t, self) == 0x00, "");
@@ -67,6 +66,7 @@ static textstartup void _main_thread_init(void) {
   // Set FS
   __install_tls((char *)td);
   assert(cthread_self()->tid == gettid());
+  __threaded = true;
 }
 
 const void *const _main_thread_ctor[] initarray = {

libc/thread/join.c

@@ -27,8 +27,7 @@
 #include "libc/str/str.h"
 #include "libc/sysv/consts/futex.h"
 #include "libc/sysv/consts/nr.h"
-#include "libc/thread/descriptor.h"
-#include "libc/thread/join.h"
+#include "libc/thread/thread.h"
 
 /**
  * Waits for thread to terminate and frees its memory.
@@ -43,7 +42,7 @@
  * @threadsafe
  */
 int cthread_join(cthread_t td, void **exitcode) {
-  int rc, tid;
+  int x, rc, tid;
   // otherwise, tid could be set to 0 even though `state` is not
   // finished mark thread as joining
   if (!td || (IsAsan() && !__asan_is_valid(td, sizeof(*td)))) {
@@ -55,11 +54,10 @@ int cthread_join(cthread_t td, void **exitcode) {
     rc = EINVAL;
   } else {
     if (~atomic_fetch_add(&td->state, cthread_joining) & cthread_finished) {
-      if (IsLinux() || IsOpenbsd()) {
+      while ((x = atomic_load(&td->tid))) {
         // FUTEX_WAIT_PRIVATE makes it hang
-        futex((uint32_t *)&td->tid, FUTEX_WAIT, tid, 0, 0);
+        cthread_memory_wait32((uint32_t *)&td->tid, x, 0);
       }
-      _spinlock(&td->tid);
     }
     if (exitcode) {
       *exitcode = td->exitcode;

libc/thread/join.h

@@ -1,11 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_JOIN_H_
-#define COSMOPOLITAN_LIBC_THREAD_JOIN_H_
-#include "libc/thread/descriptor.h"
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-int cthread_join(cthread_t, void **);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_JOIN_H_ */

libc/thread/self.c

@@ -16,7 +16,8 @@
 │ TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR             │
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
-#include "libc/thread/self.h"
+#include "libc/nexgen32e/threaded.h"
+#include "libc/thread/thread.h"
 
 STATIC_YOINK("_main_thread_ctor");
 
@@ -24,5 +25,5 @@ STATIC_YOINK("_main_thread_ctor");
  * Returns thread descriptor of the current thread.
  */
 cthread_t(cthread_self)(void) {
-  return cthread_self();
+  return (cthread_t)__get_tls_inline();
 }

libc/thread/self.h

@@ -1,22 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_SELF_H_
-#define COSMOPOLITAN_LIBC_THREAD_SELF_H_
-#include "libc/nexgen32e/threaded.h"
-#include "libc/thread/descriptor.h"
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-extern const void *const _main_thread_ctor[];
-
-#if defined(__GNUC__) && !defined(__STRICT_ANSI__)
-#define cthread_self()        \
-  ({                          \
-    YOINK(_main_thread_ctor); \
-    ((cthread_t)__get_tls()); \
-  })
-#else
-cthread_t cthread_self(void);
-#endif
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_SELF_H_ */

libc/thread/sem.c

@@ -17,9 +17,7 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/bits/atomic.h"
-#include "libc/thread/sem.h"
-#include "libc/thread/wait.h"
-#include "libc/thread/yield.h"
+#include "libc/thread/thread.h"
 
 STATIC_YOINK("_main_thread_ctor");
 

libc/thread/sem.h

@@ -1,24 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_NATIVESEM_H_
-#define COSMOPOLITAN_LIBC_THREAD_NATIVESEM_H_
-#include "libc/calls/struct/timespec.h"
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-/**
- * @fileoverview native semaphore for implementation details
- */
-
-typedef union cthread_sem_t {
-  struct {
-    uint64_t count;
-  } linux;
-} cthread_sem_t;
-
-int cthread_sem_init(cthread_sem_t *, int);
-int cthread_sem_destroy(cthread_sem_t *);
-int cthread_sem_wait(cthread_sem_t *, int, const struct timespec *);
-int cthread_sem_signal(cthread_sem_t *);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_SELF_H_ */

libc/thread/thread.h

@@ -0,0 +1,80 @@
+#ifndef COSMOPOLITAN_LIBC_THREAD_THREAD_H_
+#define COSMOPOLITAN_LIBC_THREAD_THREAD_H_
+#include "libc/calls/struct/timespec.h"
+#include "libc/runtime/runtime.h"
+
+#define CTHREAD_CREATE_DETACHED 1
+#define CTHREAD_CREATE_JOINABLE 0
+
+#if !(__ASSEMBLER__ + __LINKER__ + 0)
+COSMOPOLITAN_C_START_
+
+enum cthread_state {
+  cthread_started = 0,
+  cthread_joining = 1,
+  cthread_finished = 2,
+  cthread_detached = 4,
+  cthread_main = 127,
+};
+
+struct cthread_descriptor_t {
+  struct cthread_descriptor_t *self;  /* 0x00 */
+  void *(*func)(void *);              /* 0x08 */
+  int32_t __pad0;                     /* 0x10 */
+  int32_t state;                      /* 0x14 */
+  void *arg;                          /* 0x18 */
+  void *pthread_ret_ptr;              /* 0x20 */
+  int64_t __pad1;                     /* 0x28 */
+  struct cthread_descriptor_t *self2; /* 0x30 */
+  int32_t tid;                        /* 0x38 */
+  int32_t err;                        /* 0x3c */
+  void *exitcode;
+  struct {
+    char *top, *bottom;
+  } stack, alloc;
+  jmp_buf exiter;
+};
+
+typedef struct cthread_descriptor_t *cthread_t;
+
+typedef union cthread_sem_t {
+  struct {
+    uint64_t count;
+  } linux;
+} cthread_sem_t;
+
+typedef struct cthread_attr_t {
+  size_t stacksize, guardsize;
+  int mode;
+} cthread_attr_t;
+
+extern const void *const _main_thread_ctor[];
+
+int cthread_create(cthread_t *, const cthread_attr_t *, void *(*)(void *),
+                   void *);
+
+int cthread_yield(void);
+cthread_t cthread_self(void);
+int cthread_join(cthread_t, void **);
+void cthread_exit(void *) wontreturn;
+int cthread_detach(cthread_t);
+int cthread_attr_init(cthread_attr_t *);
+int cthread_attr_destroy(cthread_attr_t *);
+int cthread_attr_setstacksize(cthread_attr_t *, size_t);
+size_t thread_attr_getstacksize(const cthread_attr_t *);
+int cthread_attr_setguardsize(cthread_attr_t *, size_t);
+size_t cthread_attr_getguardsize(const cthread_attr_t *);
+int cthread_attr_setdetachstate(cthread_attr_t *, int);
+int cthread_attr_getdetachstate(const cthread_attr_t *);
+int cthread_sem_init(cthread_sem_t *, int);
+int cthread_sem_destroy(cthread_sem_t *);
+int cthread_sem_wait(cthread_sem_t *, int, const struct timespec *);
+int cthread_sem_signal(cthread_sem_t *);
+int cthread_memory_wait32(uint32_t *, uint32_t, const struct timespec *);
+int cthread_memory_wake32(uint32_t *, int);
+void cthread_zombies_add(cthread_t);
+void cthread_zombies_reap(void);
+
+COSMOPOLITAN_C_END_
+#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
+#endif /* COSMOPOLITAN_LIBC_THREAD_THREAD_H_ */

libc/thread/wait.c

@@ -19,12 +19,33 @@
 #include "libc/bits/atomic.h"
 #include "libc/calls/calls.h"
 #include "libc/dce.h"
+#include "libc/sysv/consts/clock.h"
 #include "libc/sysv/consts/futex.h"
+#include "libc/thread/freebsd.internal.h"
+#include "libc/thread/thread.h"
 
 int cthread_memory_wait32(uint32_t* addr, uint32_t val,
                           const struct timespec* timeout) {
+  size_t size;
+  struct _umtx_time *put, ut;
   if (IsLinux() || IsOpenbsd()) {
     return futex(addr, FUTEX_WAIT, val, timeout, 0);
+
+#if 0
+  } else if (IsFreebsd()) {
+    if (!timeout) {
+      put = 0;
+      size = 0;
+    } else {
+      ut._flags = 0;
+      ut._clockid = CLOCK_REALTIME;
+      ut._timeout = *timeout;
+      put = &ut;
+      size = sizeof(ut);
+    }
+    return _umtx_op(addr, UMTX_OP_MUTEX_WAIT, 0, &size, put);
+#endif
+
   } else {
     unsigned tries;
     for (tries = 1; atomic_load(addr) == val; ++tries) {
@@ -41,6 +62,10 @@ int cthread_memory_wait32(uint32_t* addr, uint32_t val,
 int cthread_memory_wake32(uint32_t* addr, int n) {
   if (IsLinux() || IsOpenbsd()) {
     return futex(addr, FUTEX_WAKE, n, 0, 0);
+#if 0
+  } else if (IsFreebsd()) {
+    return _umtx_op(addr, UMTX_OP_MUTEX_WAKE, n, 0, 0);
+#endif
   }
   return -1;
 }

libc/thread/wait.h

@@ -1,16 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_WAIT_H_
-#define COSMOPOLITAN_LIBC_THREAD_WAIT_H_
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-/**
- * @fileoverview wait on memory
- */
-struct timespec;
-
-int cthread_memory_wait32(uint32_t*, uint32_t, const struct timespec*);
-int cthread_memory_wake32(uint32_t*, int);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_WAIT_H_ */

libc/thread/yield.c

@@ -17,7 +17,7 @@
 │ PERFORMANCE OF THIS SOFTWARE.                                                │
 ╚─────────────────────────────────────────────────────────────────────────────*/
 #include "libc/calls/calls.h"
-#include "libc/thread/yield.h"
+#include "libc/thread/thread.h"
 
 /**
  * Asks operating system to handoff remaining time slice.

libc/thread/yield.h

@@ -1,14 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_YIELD_H_
-#define COSMOPOLITAN_LIBC_THREAD_YIELD_H_
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-/**
- * @fileoverview yield thread to OS scheduler
- */
-
-int cthread_yield(void);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_YIELD_H_ */ 

libc/thread/zombie.c

@@ -19,7 +19,7 @@
 #include "libc/bits/atomic.h"
 #include "libc/mem/mem.h"
 #include "libc/runtime/runtime.h"
-#include "libc/thread/zombie.h"
+#include "libc/thread/thread.h"
 
 static struct Zombie {
   struct Zombie *next;

libc/thread/zombie.h

@@ -1,12 +0,0 @@
-#ifndef COSMOPOLITAN_LIBC_THREAD_ZOMBIE_H_
-#define COSMOPOLITAN_LIBC_THREAD_ZOMBIE_H_
-#include "libc/thread/descriptor.h"
-#if !(__ASSEMBLER__ + __LINKER__ + 0)
-COSMOPOLITAN_C_START_
-
-void cthread_zombies_add(cthread_t);
-void cthread_zombies_reap(void);
-
-COSMOPOLITAN_C_END_
-#endif /* !(__ASSEMBLER__ + __LINKER__ + 0) */
-#endif /* COSMOPOLITAN_LIBC_THREAD_ZOMBIE_H_ */