diff --git a/liblfi/include/lfi.h b/liblfi/include/lfi.h
index c08ba3b..9bfd054 100644
--- a/liblfi/include/lfi.h
+++ b/liblfi/include/lfi.h
@@ -95,6 +95,6 @@ struct LFILoadOpts {
 bool                    lfi_proc_loadelf(struct LFIAddrSpace* as, uint8_t* prog, size_t progsz, uint8_t* interp, size_t interpsz, struct LFILoadInfo* o_info, struct LFILoadOpts opts);
 bool                    lfi_proc_init(struct LFIContext* ctx, struct LFIAddrSpace* as, struct LFILoadInfo info);
 
-void lfi_thread_init(void* thread_create, void* pausefn);
+void lfi_thread_init(void (*thread_create)(void*), void* pausefn);
 
 char* lfi_strerror(void);
diff --git a/liblfi/pal/ctx.c b/liblfi/pal/ctx.c
index 2faef6f..a74ad77 100644
--- a/liblfi/pal/ctx.c
+++ b/liblfi/pal/ctx.c
@@ -158,13 +158,13 @@ pal_register_clonectx(struct LFIContext* ctx)
 }
 
 EXPORT void
-lfi_thread_init(void* thread_create, void* pausefn)
+lfi_thread_init(void (*thread_create)(void*), void* pausefn)
 {
     // invoke sbx_thread_create(&_lfi_pause) with the clone context
     LOCK_WITH_DEFER(&lfi_clonectx_lk, lk);
-    *lfi_regs_entry(&lfi_clonectx->regs) = (uintptr_t) thread_create;
-    *lfi_regs_arg0(&lfi_clonectx->regs) = (uintptr_t) pausefn;
-    lfi_ctx_run(lfi_clonectx, lfi_clonectx->as);
+    assert(lfi_myctx == NULL);
+    lfi_myctx = lfi_clonectx;
+    thread_create(pausefn);
     lfi_myctx = lfi_newctx;
     lfi_newctx = NULL;
 }
diff --git a/liblfi/syscalls/sys_thread.c b/liblfi/syscalls/sys_thread.c
index d38d7ad..6b6f553 100644
--- a/liblfi/syscalls/sys_thread.c
+++ b/liblfi/syscalls/sys_thread.c
@@ -131,12 +131,15 @@ spawn(struct TuxThread* p, uint64_t flags, uint64_t stack, uint64_t ptidp, uint6
         // of creating a new kernel thread, we just save the stack and tls that
         // was created so it can be reused when we need to spawn threads in the
         // future.
+        struct LFIContext* save_ctx = lfi_myctx;
+        threadspawn(p2);
+        lfi_myctx = save_ctx;
         pal_register_clonectx(p2->p_ctx);
     } else if (p->p_ctx == lfi_clonectx) {
+        struct LFIContext* save_ctx = lfi_myctx;
         threadspawn(p2);
+        lfi_myctx = save_ctx;
         lfi_newctx = p2->p_ctx;
-        // does not return
-        lfi_ctx_pause(p->p_ctx, 0);
     } else {
         // Actually create a new thread.
         pthread_t thread;