diff --git a/src/runtime/os_windows.go b/src/runtime/os_windows.go
index e52f9e6bd53..576988e1077 100644
--- a/src/runtime/os_windows.go
+++ b/src/runtime/os_windows.go
@@ -955,6 +955,22 @@ func mdestroy(mp *m) {
 	}
 }
 
+// asmstdcall_trampoline calls asmstdcall converting from Go to C calling convention.
+func asmstdcall_trampoline(args unsafe.Pointer)
+
+// stdcall_no_g calls asmstdcall on os stack without using g.
+//
+//go:nosplit
+func stdcall_no_g(fn stdFunction, n int, args uintptr) uintptr {
+	libcall := libcall{
+		fn:   uintptr(unsafe.Pointer(fn)),
+		n:    uintptr(n),
+		args: args,
+	}
+	asmstdcall_trampoline(noescape(unsafe.Pointer(&libcall)))
+	return libcall.r1
+}
+
 // Calling stdcall on os stack.
 // May run during STW, so write barriers are not allowed.
 //
@@ -1054,16 +1070,17 @@ func stdcall7(fn stdFunction, a0, a1, a2, a3, a4, a5, a6 uintptr) uintptr {
 
 // These must run on the system stack only.
 func usleep2(dt int32)
-func switchtothread()
 
 //go:nosplit
 func osyield_no_g() {
-	switchtothread()
+	stdcall_no_g(_SwitchToThread, 0, 0)
 }
 
 //go:nosplit
 func osyield() {
-	systemstack(switchtothread)
+	systemstack(func() {
+		stdcall0(_SwitchToThread)
+	})
 }
 
 //go:nosplit
diff --git a/src/runtime/sys_windows_386.s b/src/runtime/sys_windows_386.s
index 2efde4515be..c8d4bf93a7a 100644
--- a/src/runtime/sys_windows_386.s
+++ b/src/runtime/sys_windows_386.s
@@ -11,6 +11,9 @@
 #define TEB_TlsSlots 0xE10
 #define TEB_ArbitraryPtr 0x14
 
+TEXT runtime·asmstdcall_trampoline<ABIInternal>(SB),NOSPLIT,$0
+	JMP	runtime·asmstdcall(SB)
+
 // void runtime·asmstdcall(void *c);
 TEXT runtime·asmstdcall(SB),NOSPLIT,$0
 	MOVL	fn+0(FP), BX
@@ -248,14 +251,6 @@ TEXT runtime·usleep2(SB),NOSPLIT,$20-4
 	MOVL	BP, SP
 	RET
 
-// Runs on OS stack.
-TEXT runtime·switchtothread(SB),NOSPLIT,$0
-	MOVL	SP, BP
-	MOVL	runtime·_SwitchToThread(SB), AX
-	CALL	AX
-	MOVL	BP, SP
-	RET
-
 TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
 loop:
 	MOVL	(_INTERRUPT_TIME+time_hi1), AX
diff --git a/src/runtime/sys_windows_amd64.s b/src/runtime/sys_windows_amd64.s
index 7a7905e56a3..af2b52735d9 100644
--- a/src/runtime/sys_windows_amd64.s
+++ b/src/runtime/sys_windows_amd64.s
@@ -12,6 +12,10 @@
 #define TEB_TlsSlots 0x1480
 #define TEB_ArbitraryPtr 0x28
 
+TEXT runtime·asmstdcall_trampoline<ABIInternal>(SB),NOSPLIT,$0
+	MOVQ	AX, CX
+	JMP	runtime·asmstdcall(SB)
+
 // void runtime·asmstdcall(void *c);
 TEXT runtime·asmstdcall(SB),NOSPLIT,$16
 	MOVQ	SP, AX
@@ -258,18 +262,6 @@ TEXT runtime·usleep2(SB),NOSPLIT,$48-4
 	MOVQ	40(SP), SP
 	RET
 
-// Runs on OS stack.
-TEXT runtime·switchtothread(SB),NOSPLIT,$0
-	MOVQ	SP, AX
-	ANDQ	$~15, SP	// alignment as per Windows requirement
-	SUBQ	$(48), SP	// room for SP and 4 args as per Windows requirement
-				// plus one extra word to keep stack 16 bytes aligned
-	MOVQ	AX, 32(SP)
-	MOVQ	runtime·_SwitchToThread(SB), AX
-	CALL	AX
-	MOVQ	32(SP), SP
-	RET
-
 TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
 	MOVQ	$_INTERRUPT_TIME, DI
 	MOVQ	time_lo(DI), AX
diff --git a/src/runtime/sys_windows_arm.s b/src/runtime/sys_windows_arm.s
index c9fca19981b..4b941e40ae5 100644
--- a/src/runtime/sys_windows_arm.s
+++ b/src/runtime/sys_windows_arm.s
@@ -9,6 +9,9 @@
 
 // Note: For system ABI, R0-R3 are args, R4-R11 are callee-save.
 
+TEXT runtime·asmstdcall_trampoline<ABIInternal>(SB),NOSPLIT,$0
+	B	runtime·asmstdcall(SB)
+
 // void runtime·asmstdcall(void *c);
 TEXT runtime·asmstdcall(SB),NOSPLIT|NOFRAME,$0
 	MOVM.DB.W [R4, R5, R14], (R13)	// push {r4, r5, lr}
@@ -211,16 +214,6 @@ TEXT runtime·usleep2(SB),NOSPLIT|NOFRAME,$0-4
 	MOVW	R4, R13			// Restore SP
 	MOVM.IA.W (R13), [R4, R15]	// pop {R4, pc}
 
-// Runs on OS stack.
-TEXT runtime·switchtothread(SB),NOSPLIT|NOFRAME,$0
-	MOVM.DB.W [R4, R14], (R13)  	// push {R4, lr}
-	MOVW    R13, R4
-	BIC	$0x7, R13		// alignment for ABI
-	MOVW	runtime·_SwitchToThread(SB), R0
-	BL	(R0)
-	MOVW 	R4, R13			// restore stack pointer
-	MOVM.IA.W (R13), [R4, R15]	// pop {R4, pc}
-
 TEXT ·publicationBarrier(SB),NOSPLIT|NOFRAME,$0-0
 	B	runtime·armPublicationBarrier(SB)
 
diff --git a/src/runtime/sys_windows_arm64.s b/src/runtime/sys_windows_arm64.s
index 2781c3c7d09..4818994ed48 100644
--- a/src/runtime/sys_windows_arm64.s
+++ b/src/runtime/sys_windows_arm64.s
@@ -19,6 +19,9 @@
 //
 // load_g and save_g (in tls_arm64.s) clobber R27 (REGTMP) and R0.
 
+TEXT runtime·asmstdcall_trampoline<ABIInternal>(SB),NOSPLIT,$0
+	B	runtime·asmstdcall(SB)
+
 // void runtime·asmstdcall(void *c);
 TEXT runtime·asmstdcall(SB),NOSPLIT,$16
 	STP	(R19, R20), 16(RSP) // save old R19, R20
@@ -240,14 +243,6 @@ TEXT runtime·usleep2(SB),NOSPLIT,$32-4
 	ADD	$16, RSP
 	RET
 
-// Runs on OS stack.
-TEXT runtime·switchtothread(SB),NOSPLIT,$16-0
-	MOVD	runtime·_SwitchToThread(SB), R0
-	SUB	$16, RSP	// skip over saved frame pointer below RSP
-	BL	(R0)
-	ADD	$16, RSP
-	RET
-
 TEXT runtime·nanotime1(SB),NOSPLIT,$0-8
 	MOVD	$_INTERRUPT_TIME, R3
 	MOVD	time_lo(R3), R0