Re-correct TSAN integration with task switching

This reverts commit 15772ba
"Update references to tsan state (#42440)", and fixes integration.

Looks like #36929 got reverted
when trying to simplify the code, but it now isn't legal. Since these
must be inlined in the right point in the runtime call/return context,
use a macro to ensure that.

src/gc-stacks.c

@@ -234,9 +234,9 @@ void sweep_stack_pools(void)
                     _jl_free_stack(ptls2, stkbuf, bufsz);
                 }
 #ifdef _COMPILER_TSAN_ENABLED_
-                if (t->tsan_state) {
-                    __tsan_destroy_fiber(t->tsan_state);
-                    t->tsan_state = NULL;
+                if (t->ctx.tsan_state) {
+                    __tsan_destroy_fiber(t->ctx.tsan_state);
+                    t->ctx.tsan_state = NULL;
                 }
 #endif
             }

src/julia.h

@@ -1846,18 +1846,8 @@ typedef struct _jl_task_t {
     jl_excstack_t *excstack;
     // current exception handler
     jl_handler_t *eh;
-
-    union {
-        jl_ucontext_t ctx; // saved thread state
-#ifdef _OS_WINDOWS_
-        jl_ucontext_t copy_stack_ctx;
-#else
-        struct jl_stack_context_t copy_stack_ctx;
-#endif
-    };
-#if defined(_COMPILER_TSAN_ENABLED_)
-    void *tsan_state;
-#endif
+    // saved thread state
+    jl_ucontext_t ctx;
     void *stkbuf; // malloc'd memory (either copybuf or stack)
     size_t bufsz; // actual sizeof stkbuf
     unsigned int copy_stack:31; // sizeof stack for copybuf

src/julia_threads.h

@@ -32,8 +32,12 @@ JL_DLLEXPORT void jl_threading_profile(void);
 
 #ifdef _OS_WINDOWS_
 #define JL_HAVE_UCONTEXT
-typedef win32_ucontext_t jl_ucontext_t;
+typedef win32_ucontext_t jl_stack_context_t;
+typedef jl_stack_context_t _jl_ucontext_t;
 #else
+typedef struct {
+    jl_jmp_buf uc_mcontext;
+} jl_stack_context_t;
 #if !defined(JL_HAVE_UCONTEXT) && \
     !defined(JL_HAVE_ASM) && \
     !defined(JL_HAVE_UNW_CONTEXT) && \
@@ -56,13 +60,8 @@ typedef win32_ucontext_t jl_ucontext_t;
 #endif
 #endif
 
-
-struct jl_stack_context_t {
-    jl_jmp_buf uc_mcontext;
-};
-
 #if (!defined(JL_HAVE_UNW_CONTEXT) && defined(JL_HAVE_ASM)) || defined(JL_HAVE_SIGALTSTACK)
-typedef struct jl_stack_context_t jl_ucontext_t;
+typedef jl_stack_context_t _jl_ucontext_t;
 #endif
 #if defined(JL_HAVE_ASYNCIFY)
 #if defined(_COMPILER_TSAN_ENABLED_)
@@ -75,19 +74,30 @@ typedef struct {
     // __asyncify_data struct.
     void *stackbottom;
     void *stacktop;
-} jl_ucontext_t;
+} _jl_ucontext_t;
 #endif
 #if defined(JL_HAVE_UNW_CONTEXT)
 #define UNW_LOCAL_ONLY
 #include <libunwind.h>
-typedef unw_context_t jl_ucontext_t;
+typedef unw_context_t _jl_ucontext_t;
 #endif
 #if defined(JL_HAVE_UCONTEXT)
 #include <ucontext.h>
-typedef ucontext_t jl_ucontext_t;
+typedef ucontext_t _jl_ucontext_t;
 #endif
 #endif
 
+typedef struct {
+    union {
+        _jl_ucontext_t ctx;
+        jl_stack_context_t copy_ctx;
+    };
+#if defined(_COMPILER_TSAN_ENABLED_)
+    void *tsan_state;
+#endif
+} jl_ucontext_t;
+
+
 // handle to reference an OS thread
 #ifdef _OS_WINDOWS_
 typedef DWORD jl_thread_t;
@@ -225,13 +235,9 @@ typedef struct _jl_tls_states_t {
     void *stackbase;
     size_t stacksize;
     union {
-        jl_ucontext_t base_ctx; // base context of stack
+        _jl_ucontext_t base_ctx; // base context of stack
         // This hack is needed to support always_copy_stacks:
-#ifdef _OS_WINDOWS_
-        jl_ucontext_t copy_stack_ctx;
-#else
-        struct jl_stack_context_t copy_stack_ctx;
-#endif
+        jl_stack_context_t copy_stack_ctx;
     };
     // Temp storage for exception thrown in signal handler. Not rooted.
     struct _jl_value_t *sig_exception;

src/signals-unix.c

@@ -44,7 +44,8 @@
 #include "julia_assert.h"
 
 // helper function for returning the unw_context_t inside a ucontext_t
-static bt_context_t *jl_to_bt_context(void *sigctx)
+// (also used by stackwalk.c)
+bt_context_t *jl_to_bt_context(void *sigctx)
 {
 #ifdef __APPLE__
     return (bt_context_t*)&((ucontext64_t*)sigctx)->uc_mcontext64->__ss;

src/signals-win.c

@@ -59,7 +59,6 @@ static void jl_try_throw_sigint(void)
 
 void __cdecl crt_sig_handler(int sig, int num)
 {
-    jl_task_t *ct = jl_current_task;
     CONTEXT Context;
     switch (sig) {
     case SIGFPE:
@@ -100,8 +99,8 @@ void __cdecl crt_sig_handler(int sig, int num)
 // StackOverflowException needs extra stack space to record the backtrace
 // so we keep one around, shared by all threads
 static jl_mutex_t backtrace_lock;
-static jl_ucontext_t collect_backtrace_fiber;
-static jl_ucontext_t error_return_fiber;
+static win32_ucontext_t collect_backtrace_fiber;
+static win32_ucontext_t error_return_fiber;
 static PCONTEXT stkerror_ctx;
 static jl_ptls_t stkerror_ptls;
 static int have_backtrace_fiber;

src/stackwalk.c

@@ -717,7 +717,7 @@ void jl_rec_backtrace(jl_task_t *t)
 #if defined(_OS_WINDOWS_)
     bt_context_t c;
     memset(&c, 0, sizeof(c));
-    _JUMP_BUFFER *mctx = (_JUMP_BUFFER*)&t->ctx.uc_mcontext;
+    _JUMP_BUFFER *mctx = (_JUMP_BUFFER*)&t->ctx.ctx.uc_mcontext;
 #if defined(_CPU_X86_64_)
     c.Rbx = mctx->Rbx;
     c.Rsp = mctx->Rsp;
@@ -737,9 +737,9 @@ void jl_rec_backtrace(jl_task_t *t)
 #endif
     context = &c;
 #elif defined(JL_HAVE_UNW_CONTEXT)
-    context = &t->ctx;
+    context = &t->ctx.ctx;
 #elif defined(JL_HAVE_UCONTEXT)
-    context = jl_to_bt_context(&t->ctx);
+    context = jl_to_bt_context(&t->ctx.ctx);
 #else
 #endif
     if (context)