runtime/cgo: store M for C-created thread in pthread key

This reapplies CL 485500, with a fix drafted in CL 492987 incorporated.

CL 485500 is reverted due to #60004 and #60007. #60004 is fixed in
CL 492743. #60007 is fixed in CL 492987 (incorporated in this CL).

[Original CL 485500 description]

This reapplies CL 481061, with the followup fixes in CL 482975, CL 485315, and
CL 485316 incorporated.

CL 481061, by doujiang24 <doujiang24@gmail.com>, speed up C to Go
calls by binding the M to the C thread. See below for its
description.
CL 482975 is a followup fix to a C declaration in testprogcgo.
CL 485315 is a followup fix for x_cgo_getstackbound on Illumos.
CL 485316 is a followup cleanup for ppc64 assembly.

CL 479915 passed the G to _cgo_getstackbound for direct updates to
gp.stack.lo. A G can be reused on a new thread after the previous thread
exited. This could trigger the C TSAN race detector because it couldn't
see the synchronization in Go (lockextra) preventing the same G from
being used on multiple threads at the same time.

We work around this by passing the address of a stack variable to
_cgo_getstackbound rather than the G. The stack is generally unique per
thread, so TSAN won't see the same address from multiple threads. Even
if stacks are reused across threads by pthread, C TSAN should see the
synchonization in the stack allocator.

A regression test is added to misc/cgo/testsanitizer.

[Original CL 481061 description]

This reapplies CL 392854, with the followup fixes in CL 479255,
CL 479915, and CL 481057 incorporated.

CL 392854, by doujiang24 <doujiang24@gmail.com>, speed up C to Go
calls by binding the M to the C thread. See below for its
description.
CL 479255 is a followup fix for a small bug in ARM assembly code.
CL 479915 is another followup fix to address C to Go calls after
the C code uses some stack, but that CL is also buggy.
CL 481057, by Michael Knyszek, is a followup fix for a memory leak
bug of CL 479915.

[Original CL 392854 description]

In a C thread, it's necessary to acquire an extra M by using needm while invoking a Go function from C. But, needm and dropm are heavy costs due to the signal-related syscalls.
So, we change to not dropm while returning back to C, which means binding the extra M to the C thread until it exits, to avoid needm and dropm on each C to Go call.
Instead, we only dropm while the C thread exits, so the extra M won't leak.

When invoking a Go function from C:
Allocate a pthread variable using pthread_key_create, only once per shared object, and register a thread-exit-time destructor.
And store the g0 of the current m into the thread-specified value of the pthread key,  only once per C thread, so that the destructor will put the extra M back onto the extra M list while the C thread exits.

When returning back to C:
Skip dropm in cgocallback, when the pthread variable has been created, so that the extra M will be reused the next time invoke a Go function from C.

This is purely a performance optimization. The old version, in which needm & dropm happen on each cgo call, is still correct too, and we have to keep the old version on systems with cgo but without pthreads, like Windows.

This optimization is significant, and the specific value depends on the OS system and CPU, but in general, it can be considered as 10x faster, for a simple Go function call from a C thread.

For the newly added BenchmarkCGoInCThread, some benchmark results:
1. it's 28x faster, from 3395 ns/op to 121 ns/op, in darwin OS & Intel(R) Core(TM) i7-9750H CPU @ 2.60GHz
2. it's 6.5x faster, from 1495 ns/op to 230 ns/op, in Linux OS & Intel(R) Xeon(R) CPU E5-2630 0 @ 2.30GHz

[CL 479915 description]

Currently, when C calls into Go the first time, we grab an M
using needm, which sets m.g0's stack bounds using the SP. We don't
know how big the stack is, so we simply assume 32K. Previously,
when the Go function returns to C, we drop the M, and the next
time C calls into Go, we put a new stack bound on the g0 based on
the current SP. After CL 392854, we don't drop the M, and the next
time C calls into Go, we reuse the same g0, without recomputing
the stack bounds. If the C code uses quite a bit of stack space
before calling into Go, the SP may be well below the 32K stack
bound we assumed, so the runtime thinks the g0 stack overflows.

This CL makes needm get a more accurate stack bound from
pthread. (In some platforms this may still be a guess as we don't
know exactly where we are in the C stack), but it is probably
better than simply assuming 32K.

[CL 492987 description]

On the first call into Go from a C thread, currently we set the g0
stack's high bound imprecisely based on the SP. With CL 485500, we
keep the M and don't recompute the stack bounds when it calls into
Go again. If the first call is made when the C thread uses some
deep stack, but a subsequent call is made with a shallower stack,
the SP may be above g0.stack.hi.

This is usually okay as we don't check usually stack.hi. One place
where we do check for stack.hi is in the signal handler, in
adjustSignalStack. In particular, C TSAN delivers signals on the
g0 stack (instead of the usual signal stack). If the SP is above
g0.stack.hi, we don't see it is on the g0 stack, and throws.

This CL makes it get an accurate stack upper bound with the
pthread API (on the platforms where it is available).

Also add some debug print for the "handler not on signal stack"
throw.

Fixes #51676.
Fixes #59294.
Fixes #59678.
Fixes #60007.

Change-Id: Ie51c8e81ade34ec81d69fd7bce1fe0039a470776
Reviewed-on: https://go-review.googlesource.com/c/go/+/495855
Run-TryBot: Cherry Mui <cherryyz@google.com>
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Michael Pratt <mpratt@google.com>

46 files changed, 1018 insertions, 69 deletions

diff --git a/src/cmd/cgo/internal/test/cgo_test.go b/src/cmd/cgo/internal/test/cgo_test.go
index 5a07c4c0fa..5e02888b3d 100644
--- a/src/cmd/cgo/internal/test/cgo_test.go
+++ b/src/cmd/cgo/internal/test/cgo_test.go
@@ -106,6 +106,7 @@ func TestThreadLock(t *testing.T)            { testThreadLockFunc(t) }
 func TestUnsignedInt(t *testing.T)           { testUnsignedInt(t) }
 func TestZeroArgCallback(t *testing.T)       { testZeroArgCallback(t) }
 
-func BenchmarkCgoCall(b *testing.B)     { benchCgoCall(b) }
-func BenchmarkGoString(b *testing.B)    { benchGoString(b) }
-func BenchmarkCGoCallback(b *testing.B) { benchCallback(b) }
+func BenchmarkCgoCall(b *testing.B)      { benchCgoCall(b) }
+func BenchmarkGoString(b *testing.B)     { benchGoString(b) }
+func BenchmarkCGoCallback(b *testing.B)  { benchCallback(b) }
+func BenchmarkCGoInCThread(b *testing.B) { benchCGoInCthread(b) }
diff --git a/src/cmd/cgo/internal/test/cthread_unix.c b/src/cmd/cgo/internal/test/cthread_unix.c
index b6ec39816b..d0da643158 100644
--- a/src/cmd/cgo/internal/test/cthread_unix.c
+++ b/src/cmd/cgo/internal/test/cthread_unix.c
@@ -32,3 +32,27 @@ doAdd(int max, int nthread)
 	for(i=0; i<nthread; i++)
 		pthread_join(thread_id[i], 0);		
 }
+
+static void*
+goDummyCallbackThread(void* p)
+{
+	int i, max;
+
+	max = *(int*)p;
+	for(i=0; i<max; i++)
+		goDummy();
+	return NULL;
+}
+
+int
+callGoInCThread(int max)
+{
+	pthread_t thread;
+
+	if (pthread_create(&thread, NULL, goDummyCallbackThread, (void*)(&max)) != 0)
+		return -1;
+	if (pthread_join(thread, NULL) != 0)
+		return -1;
+
+	return max;
+}
diff --git a/src/cmd/cgo/internal/test/cthread_windows.c b/src/cmd/cgo/internal/test/cthread_windows.c
index 3a62ddd373..4e52209dee 100644
--- a/src/cmd/cgo/internal/test/cthread_windows.c
+++ b/src/cmd/cgo/internal/test/cthread_windows.c
@@ -35,3 +35,25 @@ doAdd(int max, int nthread)
 		CloseHandle((HANDLE)thread_id[i]);
 	}
 }
+
+__stdcall
+static unsigned int
+goDummyCallbackThread(void* p)
+{
+	int i, max;
+
+	max = *(int*)p;
+	for(i=0; i<max; i++)
+		goDummy();
+	return 0;
+}
+
+int
+callGoInCThread(int max)
+{
+	uintptr_t thread_id;
+	thread_id = _beginthreadex(0, 0, goDummyCallbackThread, &max, 0, 0);
+	WaitForSingleObject((HANDLE)thread_id, INFINITE);
+	CloseHandle((HANDLE)thread_id);
+	return max;
+}
diff --git a/src/cmd/cgo/internal/test/testx.go b/src/cmd/cgo/internal/test/testx.go
index 6a8e97ddf3..0e2a51a522 100644
--- a/src/cmd/cgo/internal/test/testx.go
+++ b/src/cmd/cgo/internal/test/testx.go
@@ -24,6 +24,7 @@ import (
 /*
 // threads
 extern void doAdd(int, int);
+extern int callGoInCThread(int);
 
 // issue 1328
 void IntoC(void);
@@ -146,6 +147,10 @@ func Add(x int) {
 	*p = 2
 }
 
+//export goDummy
+func goDummy() {
+}
+
 func testCthread(t *testing.T) {
 	if (runtime.GOOS == "darwin" || runtime.GOOS == "ios") && runtime.GOARCH == "arm64" {
 		t.Skip("the iOS exec wrapper is unable to properly handle the panic from Add")
@@ -159,6 +164,15 @@ func testCthread(t *testing.T) {
 	}
 }
 
+// Benchmark measuring overhead from C to Go in a C thread.
+// Create a new C thread and invoke Go function repeatedly in the new C thread.
+func benchCGoInCthread(b *testing.B) {
+	n := C.callGoInCThread(C.int(b.N))
+	if int(n) != b.N {
+		b.Fatal("unmatch loop times")
+	}
+}
+
 // issue 1328
 
 //export BackIntoGo
diff --git a/src/cmd/cgo/internal/testcarchive/carchive_test.go b/src/cmd/cgo/internal/testcarchive/carchive_test.go
index 51a73ee77f..a92ec46c1a 100644
--- a/src/cmd/cgo/internal/testcarchive/carchive_test.go
+++ b/src/cmd/cgo/internal/testcarchive/carchive_test.go
@@ -1265,3 +1265,57 @@ func TestPreemption(t *testing.T) {
 		t.Error(err)
 	}
 }
+
+// Issue 59294. Test calling Go function from C after using some
+// stack space.
+func TestDeepStack(t *testing.T) {
+	t.Parallel()
+
+	if !testWork {
+		defer func() {
+			os.Remove("testp9" + exeSuffix)
+			os.Remove("libgo9.a")
+			os.Remove("libgo9.h")
+		}()
+	}
+
+	cmd := exec.Command("go", "build", "-buildmode=c-archive", "-o", "libgo9.a", "./libgo9")
+	out, err := cmd.CombinedOutput()
+	t.Logf("%v\n%s", cmd.Args, out)
+	if err != nil {
+		t.Fatal(err)
+	}
+	checkLineComments(t, "libgo9.h")
+	checkArchive(t, "libgo9.a")
+
+	// build with -O0 so the C compiler won't optimize out the large stack frame
+	ccArgs := append(cc, "-O0", "-o", "testp9"+exeSuffix, "main9.c", "libgo9.a")
+	out, err = exec.Command(ccArgs[0], ccArgs[1:]...).CombinedOutput()
+	t.Logf("%v\n%s", ccArgs, out)
+	if err != nil {
+		t.Fatal(err)
+	}
+
+	argv := cmdToRun("./testp9")
+	cmd = exec.Command(argv[0], argv[1:]...)
+	sb := new(strings.Builder)
+	cmd.Stdout = sb
+	cmd.Stderr = sb
+	if err := cmd.Start(); err != nil {
+		t.Fatal(err)
+	}
+
+	timer := time.AfterFunc(time.Minute,
+		func() {
+			t.Error("test program timed out")
+			cmd.Process.Kill()
+		},
+	)
+	defer timer.Stop()
+
+	err = cmd.Wait()
+	t.Logf("%v\n%s", cmd.Args, sb)
+	if err != nil {
+		t.Error(err)
+	}
+}
diff --git a/src/cmd/cgo/internal/testcarchive/testdata/libgo9/a.go b/src/cmd/cgo/internal/testcarchive/testdata/libgo9/a.go
new file mode 100644
index 0000000000..acb08d90ec
--- /dev/null
+++ b/src/cmd/cgo/internal/testcarchive/testdata/libgo9/a.go
@@ -0,0 +1,14 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+import "runtime"
+
+import "C"
+
+func main() {}
+
+//export GoF
+func GoF() { runtime.GC() }
diff --git a/src/cmd/cgo/internal/testcarchive/testdata/main9.c b/src/cmd/cgo/internal/testcarchive/testdata/main9.c
new file mode 100644
index 0000000000..95ad4dea49
--- /dev/null
+++ b/src/cmd/cgo/internal/testcarchive/testdata/main9.c
@@ -0,0 +1,24 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#include "libgo9.h"
+
+void use(int *x) { (*x)++; }
+
+void callGoFWithDeepStack() {
+	int x[10000];
+
+	use(&x[0]);
+	use(&x[9999]);
+
+	GoF();
+
+	use(&x[0]);
+	use(&x[9999]);
+}
+
+int main() {
+	GoF();                  // call GoF without using much stack
+	callGoFWithDeepStack(); // call GoF with a deep stack
+}
diff --git a/src/cmd/cgo/internal/testsanitizers/testdata/tsan14.go b/src/cmd/cgo/internal/testsanitizers/testdata/tsan14.go
new file mode 100644
index 0000000000..d594ffb5c0
--- /dev/null
+++ b/src/cmd/cgo/internal/testsanitizers/testdata/tsan14.go
@@ -0,0 +1,53 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+package main
+
+// This program failed when run under the C/C++ ThreadSanitizer.
+//
+// cgocallback on a new thread calls into runtime.needm -> _cgo_getstackbound
+// to update gp.stack.lo with the stack bounds. If the G itself is passed to
+// _cgo_getstackbound, then writes to the same G can be seen on multiple
+// threads (when the G is reused after thread exit). This would trigger TSAN.
+
+/*
+#include <pthread.h>
+
+void go_callback();
+
+static void *thr(void *arg) {
+    go_callback();
+    return 0;
+}
+
+static void foo() {
+    pthread_t th;
+    pthread_attr_t attr;
+    pthread_attr_init(&attr);
+    pthread_attr_setstacksize(&attr, 256 << 10);
+    pthread_create(&th, &attr, thr, 0);
+    pthread_join(th, 0);
+}
+*/
+import "C"
+
+import (
+	"time"
+)
+
+//export go_callback
+func go_callback() {
+}
+
+func main() {
+	for i := 0; i < 2; i++ {
+		go func() {
+			for {
+				C.foo()
+			}
+		}()
+	}
+
+	time.Sleep(1000*time.Millisecond)
+}
diff --git a/src/cmd/cgo/internal/testsanitizers/tsan_test.go b/src/cmd/cgo/internal/testsanitizers/tsan_test.go
index cb63f873f9..6f70ebfef5 100644
--- a/src/cmd/cgo/internal/testsanitizers/tsan_test.go
+++ b/src/cmd/cgo/internal/testsanitizers/tsan_test.go
@@ -49,6 +49,7 @@ func TestTSAN(t *testing.T) {
 		{src: "tsan11.go", needsRuntime: true},
 		{src: "tsan12.go", needsRuntime: true},
 		{src: "tsan13.go", needsRuntime: true},
+		{src: "tsan14.go", needsRuntime: true},
 	}
 	for _, tc := range cases {
 		tc := tc
diff --git a/src/runtime/asm_386.s b/src/runtime/asm_386.s
index febe27089f..5fd0ab9817 100644
--- a/src/runtime/asm_386.s
+++ b/src/runtime/asm_386.s
@@ -689,7 +689,20 @@ nosave:
 TEXT ·cgocallback(SB),NOSPLIT,$12-12  // Frame size must match commented places below
 	NO_LOCAL_POINTERS
 
-	// If g is nil, Go did not create the current thread.
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVL	fn+0(FP), AX
+	CMPL	AX, $0
+	JNE	loadg
+	// Restore the g from frame.
+	get_tls(CX)
+	MOVL	frame+4(FP), BX
+	MOVL	BX, g(CX)
+	JMP	dropm
+
+loadg:
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -707,9 +720,9 @@ TEXT ·cgocallback(SB),NOSPLIT,$12-12  // Frame size must match commented places
 	MOVL	BP, savedm-4(SP) // saved copy of oldm
 	JMP	havem
 needm:
-	MOVL	$runtime·needm(SB), AX
+	MOVL	$runtime·needAndBindM(SB), AX
 	CALL	AX
-	MOVL	$0, savedm-4(SP) // dropm on return
+	MOVL	$0, savedm-4(SP)
 	get_tls(CX)
 	MOVL	g(CX), BP
 	MOVL	g_m(BP), BP
@@ -784,13 +797,29 @@ havem:
 	MOVL	0(SP), AX
 	MOVL	AX, (g_sched+gobuf_sp)(SI)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVL	savedm-4(SP), DX
 	CMPL	DX, $0
-	JNE 3(PC)
+	JNE	droppedm
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVL	_cgo_pthread_key_created(SB), DX
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	CMPL	DX, $0
+	JEQ	dropm
+	CMPL	(DX), $0
+	JNE	droppedm
+
+dropm:
 	MOVL	$runtime·dropm(SB), AX
 	CALL	AX
+droppedm:
 
 	// Done!
 	RET
diff --git a/src/runtime/asm_amd64.s b/src/runtime/asm_amd64.s
index 7fb1ae2cff..7fe8528d19 100644
--- a/src/runtime/asm_amd64.s
+++ b/src/runtime/asm_amd64.s
@@ -918,7 +918,20 @@ GLOBL zeroTLS<>(SB),RODATA,$const_tlsSize
 TEXT ·cgocallback(SB),NOSPLIT,$24-24
 	NO_LOCAL_POINTERS
 
-	// If g is nil, Go did not create the current thread.
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVQ	fn+0(FP), AX
+	CMPQ	AX, $0
+	JNE	loadg
+	// Restore the g from frame.
+	get_tls(CX)
+	MOVQ	frame+8(FP), BX
+	MOVQ	BX, g(CX)
+	JMP	dropm
+
+loadg:
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one m for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -956,9 +969,9 @@ needm:
 	// a bad value in there, in case needm tries to use it.
 	XORPS	X15, X15
 	XORQ    R14, R14
-	MOVQ	$runtime·needm<ABIInternal>(SB), AX
+	MOVQ	$runtime·needAndBindM<ABIInternal>(SB), AX
 	CALL	AX
-	MOVQ	$0, savedm-8(SP) // dropm on return
+	MOVQ	$0, savedm-8(SP)
 	get_tls(CX)
 	MOVQ	g(CX), BX
 	MOVQ	g_m(BX), BX
@@ -1047,11 +1060,26 @@ havem:
 	MOVQ	0(SP), AX
 	MOVQ	AX, (g_sched+gobuf_sp)(SI)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVQ	savedm-8(SP), BX
 	CMPQ	BX, $0
 	JNE	done
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVQ	_cgo_pthread_key_created(SB), AX
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	CMPQ	AX, $0
+	JEQ	dropm
+	CMPQ	(AX), $0
+	JNE	done
+
+dropm:
 	MOVQ	$runtime·dropm(SB), AX
 	CALL	AX
 #ifdef GOOS_windows
diff --git a/src/runtime/asm_arm.s b/src/runtime/asm_arm.s
index 01621245dc..cd692e51a3 100644
--- a/src/runtime/asm_arm.s
+++ b/src/runtime/asm_arm.s
@@ -630,6 +630,16 @@ nosave:
 TEXT	·cgocallback(SB),NOSPLIT,$12-12
 	NO_LOCAL_POINTERS
 
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVW	fn+0(FP), R1
+	CMP	$0, R1
+	B.NE	loadg
+	// Restore the g from frame.
+	MOVW	frame+4(FP), g
+	B	dropm
+
+loadg:
 	// Load m and g from thread-local storage.
 #ifdef GOOS_openbsd
 	BL	runtime·load_g(SB)
@@ -639,7 +649,8 @@ TEXT	·cgocallback(SB),NOSPLIT,$12-12
 	BL.NE	runtime·load_g(SB)
 #endif
 
-	// If g is nil, Go did not create the current thread.
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -653,7 +664,7 @@ TEXT	·cgocallback(SB),NOSPLIT,$12-12
 
 needm:
 	MOVW	g, savedm-4(SP) // g is zero, so is m.
-	MOVW	$runtime·needm(SB), R0
+	MOVW	$runtime·needAndBindM(SB), R0
 	BL	(R0)
 
 	// Set m->g0->sched.sp = SP, so that if a panic happens
@@ -724,14 +735,31 @@ havem:
 	MOVW	savedsp-12(SP), R4	// must match frame size
 	MOVW	R4, (g_sched+gobuf_sp)(g)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVW	savedm-4(SP), R6
 	CMP	$0, R6
-	B.NE	3(PC)
+	B.NE	done
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVW	_cgo_pthread_key_created(SB), R6
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	CMP	$0, R6
+	B.EQ	dropm
+	MOVW	(R6), R6
+	CMP	$0, R6
+	B.NE	done
+
+dropm:
 	MOVW	$runtime·dropm(SB), R0
 	BL	(R0)
 
+done:
 	// Done!
 	RET
 
diff --git a/src/runtime/asm_arm64.s b/src/runtime/asm_arm64.s
index 6fe04a6445..5cce33d7fe 100644
--- a/src/runtime/asm_arm64.s
+++ b/src/runtime/asm_arm64.s
@@ -1015,10 +1015,20 @@ nosave:
 TEXT ·cgocallback(SB),NOSPLIT,$24-24
 	NO_LOCAL_POINTERS
 
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVD	fn+0(FP), R1
+	CBNZ	R1, loadg
+	// Restore the g from frame.
+	MOVD	frame+8(FP), g
+	B	dropm
+
+loadg:
 	// Load g from thread-local storage.
 	BL	runtime·load_g(SB)
 
-	// If g is nil, Go did not create the current thread.
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -1031,7 +1041,7 @@ TEXT ·cgocallback(SB),NOSPLIT,$24-24
 
 needm:
 	MOVD	g, savedm-8(SP) // g is zero, so is m.
-	MOVD	$runtime·needm(SB), R0
+	MOVD	$runtime·needAndBindM(SB), R0
 	BL	(R0)
 
 	// Set m->g0->sched.sp = SP, so that if a panic happens
@@ -1112,10 +1122,24 @@ havem:
 	MOVD	savedsp-16(SP), R4
 	MOVD	R4, (g_sched+gobuf_sp)(g)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVD	savedm-8(SP), R6
 	CBNZ	R6, droppedm
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVD	_cgo_pthread_key_created(SB), R6
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	CBZ	R6, dropm
+	MOVD	(R6), R6
+	CBNZ	R6, droppedm
+
+dropm:
 	MOVD	$runtime·dropm(SB), R0
 	BL	(R0)
 droppedm:
diff --git a/src/runtime/asm_loong64.s b/src/runtime/asm_loong64.s
index 6029dbc8c3..b93ad3316d 100644
--- a/src/runtime/asm_loong64.s
+++ b/src/runtime/asm_loong64.s
@@ -460,13 +460,23 @@ g0:
 TEXT ·cgocallback(SB),NOSPLIT,$24-24
 	NO_LOCAL_POINTERS
 
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVV    fn+0(FP), R5
+	BNE	R5, loadg
+	// Restore the g from frame.
+	MOVV    frame+8(FP), g
+	JMP	dropm
+
+loadg:
 	// Load m and g from thread-local storage.
 	MOVB	runtime·iscgo(SB), R19
 	BEQ	R19, nocgo
 	JAL	runtime·load_g(SB)
 nocgo:
 
-	// If g is nil, Go did not create the current thread.
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -479,7 +489,7 @@ nocgo:
 
 needm:
 	MOVV	g, savedm-8(SP) // g is zero, so is m.
-	MOVV	$runtime·needm(SB), R4
+	MOVV	$runtime·needAndBindM(SB), R4
 	JAL	(R4)
 
 	// Set m->sched.sp = SP, so that if a panic happens
@@ -551,10 +561,24 @@ havem:
 	MOVV	savedsp-24(SP), R13 // must match frame size
 	MOVV	R13, (g_sched+gobuf_sp)(g)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVV	savedm-8(SP), R12
 	BNE	R12, droppedm
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVV	_cgo_pthread_key_created(SB), R12
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	BEQ	R12, dropm
+	MOVV    (R12), R12
+	BNE	R12, droppedm
+
+dropm:
 	MOVV	$runtime·dropm(SB), R4
 	JAL	(R4)
 droppedm:
diff --git a/src/runtime/asm_mips64x.s b/src/runtime/asm_mips64x.s
index e6eb13f00a..1da90f7777 100644
--- a/src/runtime/asm_mips64x.s
+++ b/src/runtime/asm_mips64x.s
@@ -469,13 +469,23 @@ g0:
 TEXT ·cgocallback(SB),NOSPLIT,$24-24
 	NO_LOCAL_POINTERS
 
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVV	fn+0(FP), R5
+	BNE	R5, loadg
+	// Restore the g from frame.
+	MOVV	frame+8(FP), g
+	JMP	dropm
+
+loadg:
 	// Load m and g from thread-local storage.
 	MOVB	runtime·iscgo(SB), R1
 	BEQ	R1, nocgo
 	JAL	runtime·load_g(SB)
 nocgo:
 
-	// If g is nil, Go did not create the current thread.
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -488,7 +498,7 @@ nocgo:
 
 needm:
 	MOVV	g, savedm-8(SP) // g is zero, so is m.
-	MOVV	$runtime·needm(SB), R4
+	MOVV	$runtime·needAndBindM(SB), R4
 	JAL	(R4)
 
 	// Set m->sched.sp = SP, so that if a panic happens
@@ -559,10 +569,24 @@ havem:
 	MOVV	savedsp-24(SP), R2	// must match frame size
 	MOVV	R2, (g_sched+gobuf_sp)(g)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVV	savedm-8(SP), R3
 	BNE	R3, droppedm
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVV	_cgo_pthread_key_created(SB), R3
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	BEQ	R3, dropm
+	MOVV	(R3), R3
+	BNE	R3, droppedm
+
+dropm:
 	MOVV	$runtime·dropm(SB), R4
 	JAL	(R4)
 droppedm:
diff --git a/src/runtime/asm_mipsx.s b/src/runtime/asm_mipsx.s
index fc81e76354..49f96044c4 100644
--- a/src/runtime/asm_mipsx.s
+++ b/src/runtime/asm_mipsx.s
@@ -459,13 +459,23 @@ g0:
 TEXT ·cgocallback(SB),NOSPLIT,$12-12
 	NO_LOCAL_POINTERS
 
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVW	fn+0(FP), R5
+	BNE	R5, loadg
+	// Restore the g from frame.
+	MOVW	frame+4(FP), g
+	JMP	dropm
+
+loadg:
 	// Load m and g from thread-local storage.
 	MOVB	runtime·iscgo(SB), R1
 	BEQ	R1, nocgo
 	JAL	runtime·load_g(SB)
 nocgo:
 
-	// If g is nil, Go did not create the current thread.
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -478,7 +488,7 @@ nocgo:
 
 needm:
 	MOVW	g, savedm-4(SP) // g is zero, so is m.
-	MOVW	$runtime·needm(SB), R4
+	MOVW	$runtime·needAndBindM(SB), R4
 	JAL	(R4)
 
 	// Set m->sched.sp = SP, so that if a panic happens
@@ -549,10 +559,24 @@ havem:
 	MOVW	savedsp-12(SP), R2	// must match frame size
 	MOVW	R2, (g_sched+gobuf_sp)(g)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVW	savedm-4(SP), R3
 	BNE	R3, droppedm
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVW	_cgo_pthread_key_created(SB), R3
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	BEQ	R3, dropm
+	MOVW	(R3), R3
+	BNE	R3, droppedm
+
+dropm:
 	MOVW	$runtime·dropm(SB), R4
 	JAL	(R4)
 droppedm:
diff --git a/src/runtime/asm_ppc64x.s b/src/runtime/asm_ppc64x.s
index 1e17291d78..d5be18e853 100644
--- a/src/runtime/asm_ppc64x.s
+++ b/src/runtime/asm_ppc64x.s
@@ -628,6 +628,16 @@ g0:
 TEXT ·cgocallback(SB),NOSPLIT,$24-24
 	NO_LOCAL_POINTERS
 
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVD	fn+0(FP), R5
+	CMP	R5, $0
+	BNE	loadg
+	// Restore the g from frame.
+	MOVD	frame+8(FP), g
+	BR	dropm
+
+loadg:
 	// Load m and g from thread-local storage.
 	MOVBZ	runtime·iscgo(SB), R3
 	CMP	R3, $0
@@ -635,7 +645,8 @@ TEXT ·cgocallback(SB),NOSPLIT,$24-24
 	BL	runtime·load_g(SB)
 nocgo:
 
-	// If g is nil, Go did not create the current thread.
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -649,7 +660,7 @@ nocgo:
 
 needm:
 	MOVD	g, savedm-8(SP) // g is zero, so is m.
-	MOVD	$runtime·needm(SB), R12
+	MOVD	$runtime·needAndBindM(SB), R12
 	MOVD	R12, CTR
 	BL	(CTR)
 
@@ -724,11 +735,27 @@ havem:
 	MOVD	savedsp-24(SP), R4      // must match frame size
 	MOVD	R4, (g_sched+gobuf_sp)(g)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVD	savedm-8(SP), R6
 	CMP	R6, $0
 	BNE	droppedm
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVD	_cgo_pthread_key_created(SB), R6
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	CMP	R6, $0
+	BEQ	dropm
+	MOVD	(R6), R6
+	CMP	R6, $0
+	BNE	droppedm
+
+dropm:
 	MOVD	$runtime·dropm(SB), R12
 	MOVD	R12, CTR
 	BL	(CTR)
diff --git a/src/runtime/asm_riscv64.s b/src/runtime/asm_riscv64.s
index 759bae24b5..0a34a591fd 100644
--- a/src/runtime/asm_riscv64.s
+++ b/src/runtime/asm_riscv64.s
@@ -519,13 +519,23 @@ TEXT runtime·goexit(SB),NOSPLIT|NOFRAME|TOPFRAME,$0-0
 TEXT ·cgocallback(SB),NOSPLIT,$24-24
 	NO_LOCAL_POINTERS
 
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOV	fn+0(FP), X7
+	BNE	ZERO, X7, loadg
+	// Restore the g from frame.
+	MOV	frame+8(FP), g
+	JMP	dropm
+
+loadg:
 	// Load m and g from thread-local storage.
 	MOVBU	runtime·iscgo(SB), X5
 	BEQ	ZERO, X5, nocgo
 	CALL	runtime·load_g(SB)
 nocgo:
 
-	// If g is nil, Go did not create the current thread.
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -538,7 +548,7 @@ nocgo:
 
 needm:
 	MOV	g, savedm-8(SP) // g is zero, so is m.
-	MOV	$runtime·needm(SB), X6
+	MOV	$runtime·needAndBindM(SB), X6
 	JALR	RA, X6
 
 	// Set m->sched.sp = SP, so that if a panic happens
@@ -609,10 +619,24 @@ havem:
 	MOV	savedsp-24(SP), X6	// must match frame size
 	MOV	X6, (g_sched+gobuf_sp)(g)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOV	savedm-8(SP), X5
 	BNE	ZERO, X5, droppedm
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOV	_cgo_pthread_key_created(SB), X5
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	BEQ	ZERO, X5, dropm
+	MOV	(X5), X5
+	BNE	ZERO, X5, droppedm
+
+dropm:
 	MOV	$runtime·dropm(SB), X6
 	JALR	RA, X6
 droppedm:
diff --git a/src/runtime/asm_s390x.s b/src/runtime/asm_s390x.s
index d427c07de4..4c4a42e00a 100644
--- a/src/runtime/asm_s390x.s
+++ b/src/runtime/asm_s390x.s
@@ -564,13 +564,23 @@ g0:
 TEXT ·cgocallback(SB),NOSPLIT,$24-24
 	NO_LOCAL_POINTERS
 
+	// Skip cgocallbackg, just dropm when fn is nil, and frame is the saved g.
+	// It is used to dropm while thread is exiting.
+	MOVD	fn+0(FP), R1
+	CMPBNE	R1, $0, loadg
+	// Restore the g from frame.
+	MOVD	frame+8(FP), g
+	BR	dropm
+
+loadg:
 	// Load m and g from thread-local storage.
 	MOVB	runtime·iscgo(SB), R3
 	CMPBEQ	R3, $0, nocgo
 	BL	runtime·load_g(SB)
 
 nocgo:
-	// If g is nil, Go did not create the current thread.
+	// If g is nil, Go did not create the current thread,
+	// or if this thread never called into Go on pthread platforms.
 	// Call needm to obtain one for temporary use.
 	// In this case, we're running on the thread stack, so there's
 	// lots of space, but the linker doesn't know. Hide the call from
@@ -583,7 +593,7 @@ nocgo:
 
 needm:
 	MOVD	g, savedm-8(SP) // g is zero, so is m.
-	MOVD	$runtime·needm(SB), R3
+	MOVD	$runtime·needAndBindM(SB), R3
 	BL	(R3)
 
 	// Set m->sched.sp = SP, so that if a panic happens
@@ -654,10 +664,24 @@ havem:
 	MOVD	savedsp-24(SP), R4	// must match frame size
 	MOVD	R4, (g_sched+gobuf_sp)(g)
 
-	// If the m on entry was nil, we called needm above to borrow an m
-	// for the duration of the call. Since the call is over, return it with dropm.
+	// If the m on entry was nil, we called needm above to borrow an m,
+	// 1. for the duration of the call on non-pthread platforms,
+	// 2. or the duration of the C thread alive on pthread platforms.
+	// If the m on entry wasn't nil,
+	// 1. the thread might be a Go thread,
+	// 2. or it's wasn't the first call from a C thread on pthread platforms,
+	//    since the we skip dropm to resue the m in the first call.
 	MOVD	savedm-8(SP), R6
 	CMPBNE	R6, $0, droppedm
+
+	// Skip dropm to reuse it in the next call, when a pthread key has been created.
+	MOVD	_cgo_pthread_key_created(SB), R6
+	// It means cgo is disabled when _cgo_pthread_key_created is a nil pointer, need dropm.
+	CMPBEQ	R6, $0, dropm
+	MOVD	(R6), R6
+	CMPBNE	R6, $0, droppedm
+
+dropm:
 	MOVD	$runtime·dropm(SB), R3
 	BL	(R3)
 droppedm:
diff --git a/src/runtime/cgo.go b/src/runtime/cgo.go
index d90468240d..395303552c 100644
--- a/src/runtime/cgo.go
+++ b/src/runtime/cgo.go
@@ -17,6 +17,9 @@ import "unsafe"
 //go:linkname _cgo_callers _cgo_callers
 //go:linkname _cgo_set_context_function _cgo_set_context_function
 //go:linkname _cgo_yield _cgo_yield
+//go:linkname _cgo_pthread_key_created _cgo_pthread_key_created
+//go:linkname _cgo_bindm _cgo_bindm
+//go:linkname _cgo_getstackbound _cgo_getstackbound
 
 var (
 	_cgo_init                     unsafe.Pointer
@@ -26,11 +29,17 @@ var (
 	_cgo_callers                  unsafe.Pointer
 	_cgo_set_context_function     unsafe.Pointer
 	_cgo_yield                    unsafe.Pointer
+	_cgo_pthread_key_created      unsafe.Pointer
+	_cgo_bindm                    unsafe.Pointer
+	_cgo_getstackbound            unsafe.Pointer
 )
 
 // iscgo is set to true by the runtime/cgo package
 var iscgo bool
 
+// set_crosscall2 is set by the runtime/cgo package
+var set_crosscall2 func()
+
 // cgoHasExtraM is set on startup when an extra M is created for cgo.
 // The extra M must be created before any C/C++ code calls cgocallback.
 var cgoHasExtraM bool
diff --git a/src/runtime/cgo/asm_386.s b/src/runtime/cgo/asm_386.s
index 2e7e9512e2..086e20b02f 100644
--- a/src/runtime/cgo/asm_386.s
+++ b/src/runtime/cgo/asm_386.s
@@ -4,6 +4,14 @@
 
 #include "textflag.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVL	_crosscall2_ptr(SB), AX
+	MOVL	$crosscall2(SB), BX
+	MOVL	BX, (AX)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_amd64.s b/src/runtime/cgo/asm_amd64.s
index e223a6c870..f254622f23 100644
--- a/src/runtime/cgo/asm_amd64.s
+++ b/src/runtime/cgo/asm_amd64.s
@@ -5,6 +5,14 @@
 #include "textflag.h"
 #include "abi_amd64.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVQ	_crosscall2_ptr(SB), AX
+	MOVQ	$crosscall2(SB), BX
+	MOVQ	BX, (AX)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_arm.s b/src/runtime/cgo/asm_arm.s
index ea55e173c1..f7f99772a6 100644
--- a/src/runtime/cgo/asm_arm.s
+++ b/src/runtime/cgo/asm_arm.s
@@ -4,6 +4,14 @@
 
 #include "textflag.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVW	_crosscall2_ptr(SB), R1
+	MOVW	$crosscall2(SB), R2
+	MOVW	R2, (R1)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_arm64.s b/src/runtime/cgo/asm_arm64.s
index e808dedcfc..ce8909b492 100644
--- a/src/runtime/cgo/asm_arm64.s
+++ b/src/runtime/cgo/asm_arm64.s
@@ -5,6 +5,14 @@
 #include "textflag.h"
 #include "abi_arm64.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVD	_crosscall2_ptr(SB), R1
+	MOVD	$crosscall2(SB), R2
+	MOVD	R2, (R1)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_loong64.s b/src/runtime/cgo/asm_loong64.s
index aea4f8e6b9..3b514ffc4a 100644
--- a/src/runtime/cgo/asm_loong64.s
+++ b/src/runtime/cgo/asm_loong64.s
@@ -5,6 +5,14 @@
 #include "textflag.h"
 #include "abi_loong64.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVV	_crosscall2_ptr(SB), R5
+	MOVV	$crosscall2(SB), R6
+	MOVV	R6, (R5)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_mips64x.s b/src/runtime/cgo/asm_mips64x.s
index 904f781d87..0a8fbbbef0 100644
--- a/src/runtime/cgo/asm_mips64x.s
+++ b/src/runtime/cgo/asm_mips64x.s
@@ -6,6 +6,14 @@
 
 #include "textflag.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVV	_crosscall2_ptr(SB), R5
+	MOVV	$crosscall2(SB), R6
+	MOVV	R6, (R5)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_mipsx.s b/src/runtime/cgo/asm_mipsx.s
index 5e2db0b56e..a57ae97d7e 100644
--- a/src/runtime/cgo/asm_mipsx.s
+++ b/src/runtime/cgo/asm_mipsx.s
@@ -6,6 +6,14 @@
 
 #include "textflag.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVW	_crosscall2_ptr(SB), R5
+	MOVW	$crosscall2(SB), R6
+	MOVW	R6, (R5)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_ppc64x.s b/src/runtime/cgo/asm_ppc64x.s
index cba053deb7..c258c7c2a0 100644
--- a/src/runtime/cgo/asm_ppc64x.s
+++ b/src/runtime/cgo/asm_ppc64x.s
@@ -8,6 +8,25 @@
 #include "asm_ppc64x.h"
 #include "abi_ppc64x.h"
 
+#ifdef GO_PPC64X_HAS_FUNCDESC
+// crosscall2 is marked with go:cgo_export_static. On AIX, this creates and exports
+// the symbol name and descriptor as the AIX linker expects, but does not work if
+// referenced from within Go. Create and use an aliased descriptor of crosscall2
+// to workaround this.
+DEFINE_PPC64X_FUNCDESC(_crosscall2<>, crosscall2)
+#define CROSSCALL2_FPTR $_crosscall2<>(SB)
+#else
+#define CROSSCALL2_FPTR $crosscall2(SB)
+#endif
+
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVD	_crosscall2_ptr(SB), R5
+	MOVD	CROSSCALL2_FPTR, R6
+	MOVD	R6, (R5)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
@@ -27,8 +46,12 @@ TEXT crosscall2(SB),NOSPLIT|NOFRAME,$0
 
 #ifdef GO_PPC64X_HAS_FUNCDESC
 	// Load the real entry address from the first slot of the function descriptor.
+	// The first argument fn might be null, that means dropm in pthread key destructor.
+	CMP	R3, $0
+	BEQ	nil_fn
 	MOVD	8(R3), R2
 	MOVD	(R3), R3
+nil_fn:
 #endif
 	MOVD	R3, FIXED_FRAME+0(R1)	// fn unsafe.Pointer
 	MOVD	R4, FIXED_FRAME+8(R1)	// a unsafe.Pointer
diff --git a/src/runtime/cgo/asm_riscv64.s b/src/runtime/cgo/asm_riscv64.s
index 45151bf02b..08c4ed8466 100644
--- a/src/runtime/cgo/asm_riscv64.s
+++ b/src/runtime/cgo/asm_riscv64.s
@@ -4,6 +4,14 @@
 
 #include "textflag.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOV	_crosscall2_ptr(SB), X7
+	MOV	$crosscall2(SB), X8
+	MOV	X8, (X7)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_s390x.s b/src/runtime/cgo/asm_s390x.s
index 8bf16e75e2..bb0dfc1e31 100644
--- a/src/runtime/cgo/asm_s390x.s
+++ b/src/runtime/cgo/asm_s390x.s
@@ -4,6 +4,14 @@
 
 #include "textflag.h"
 
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's such a pointer chain: _crosscall2_ptr -> x_crosscall2_ptr -> crosscall2
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	MOVD	_crosscall2_ptr(SB), R1
+	MOVD	$crosscall2(SB), R2
+	MOVD	R2, (R1)
+	RET
+
 // Called by C code generated by cmd/cgo.
 // func crosscall2(fn, a unsafe.Pointer, n int32, ctxt uintptr)
 // Saves C callee-saved registers and calls cgocallback with three arguments.
diff --git a/src/runtime/cgo/asm_wasm.s b/src/runtime/cgo/asm_wasm.s
index cb140eb7b8..e7f01bdc56 100644
--- a/src/runtime/cgo/asm_wasm.s
+++ b/src/runtime/cgo/asm_wasm.s
@@ -4,5 +4,8 @@
 
 #include "textflag.h"
 
+TEXT ·set_crosscall2(SB),NOSPLIT,$0-0
+	UNDEF
+
 TEXT crosscall2(SB), NOSPLIT, $0
 	UNDEF
diff --git a/src/runtime/cgo/callbacks.go b/src/runtime/cgo/callbacks.go
index e7c8ef3e07..3c246a88b6 100644
--- a/src/runtime/cgo/callbacks.go
+++ b/src/runtime/cgo/callbacks.go
@@ -71,6 +71,42 @@ var _cgo_thread_start = &x_cgo_thread_start
 var x_cgo_sys_thread_create byte
 var _cgo_sys_thread_create = &x_cgo_sys_thread_create
 
+// Indicates whether a dummy thread key has been created or not.
+//
+// When calling go exported function from C, we register a destructor
+// callback, for a dummy thread key, by using pthread_key_create.
+
+//go:cgo_import_static x_cgo_pthread_key_created
+//go:linkname x_cgo_pthread_key_created x_cgo_pthread_key_created
+//go:linkname _cgo_pthread_key_created _cgo_pthread_key_created
+var x_cgo_pthread_key_created byte
+var _cgo_pthread_key_created = &x_cgo_pthread_key_created
+
+// Export crosscall2 to a c function pointer variable.
+// Used to dropm in pthread key destructor, while C thread is exiting.
+
+//go:cgo_import_static x_crosscall2_ptr
+//go:linkname x_crosscall2_ptr x_crosscall2_ptr
+//go:linkname _crosscall2_ptr _crosscall2_ptr
+var x_crosscall2_ptr byte
+var _crosscall2_ptr = &x_crosscall2_ptr
+
+// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+// It's for the runtime package to call at init time.
+func set_crosscall2()
+
+//go:linkname _set_crosscall2 runtime.set_crosscall2
+var _set_crosscall2 = set_crosscall2
+
+// Store the g into the thread-specific value.
+// So that pthread_key_destructor will dropm when the thread is exiting.
+
+//go:cgo_import_static x_cgo_bindm
+//go:linkname x_cgo_bindm x_cgo_bindm
+//go:linkname _cgo_bindm _cgo_bindm
+var x_cgo_bindm byte
+var _cgo_bindm = &x_cgo_bindm
+
 // Notifies that the runtime has been initialized.
 //
 // We currently block at every CGO entry point (via _cgo_wait_runtime_init_done)
@@ -105,3 +141,12 @@ var _cgo_yield unsafe.Pointer
 
 //go:cgo_export_static _cgo_topofstack
 //go:cgo_export_dynamic _cgo_topofstack
+
+// x_cgo_getstackbound gets the thread's C stack size and
+// set the G's stack bound based on the stack size.
+
+//go:cgo_import_static x_cgo_getstackbound
+//go:linkname x_cgo_getstackbound x_cgo_getstackbound
+//go:linkname _cgo_getstackbound _cgo_getstackbound
+var x_cgo_getstackbound byte
+var _cgo_getstackbound = &x_cgo_getstackbound
diff --git a/src/runtime/cgo/gcc_libinit.c b/src/runtime/cgo/gcc_libinit.c
index 57620fe4de..9676593211 100644
--- a/src/runtime/cgo/gcc_libinit.c
+++ b/src/runtime/cgo/gcc_libinit.c
@@ -17,6 +17,14 @@ static pthread_cond_t runtime_init_cond = PTHREAD_COND_INITIALIZER;
 static pthread_mutex_t runtime_init_mu = PTHREAD_MUTEX_INITIALIZER;
 static int runtime_init_done;
 
+// pthread_g is a pthread specific key, for storing the g that binded to the C thread.
+// The registered pthread_key_destructor will dropm, when the pthread-specified value g is not NULL,
+// while a C thread is exiting.
+static pthread_key_t pthread_g;
+static void pthread_key_destructor(void* g);
+uintptr_t x_cgo_pthread_key_created;
+void (*x_crosscall2_ptr)(void (*fn)(void *), void *, int, size_t);
+
 // The context function, used when tracing back C calls into Go.
 static void (*cgo_context_function)(struct context_arg*);
 
@@ -39,6 +47,12 @@ _cgo_wait_runtime_init_done(void) {
 		pthread_cond_wait(&runtime_init_cond, &runtime_init_mu);
 	}
 
+	// The key and x_cgo_pthread_key_created are for the whole program,
+	// whereas the specific and destructor is per thread.
+	if (x_cgo_pthread_key_created == 0 && pthread_key_create(&pthread_g, pthread_key_destructor) == 0) {
+		x_cgo_pthread_key_created = 1;
+	}
+
 	// TODO(iant): For the case of a new C thread calling into Go, such
 	// as when using -buildmode=c-archive, we know that Go runtime
 	// initialization is complete but we do not know that all Go init
@@ -61,6 +75,16 @@ _cgo_wait_runtime_init_done(void) {
 	return 0;
 }
 
+// Store the g into a thread-specific value associated with the pthread key pthread_g.
+// And pthread_key_destructor will dropm when the thread is exiting.
+void x_cgo_bindm(void* g) {
+	// We assume this will always succeed, otherwise, there might be extra M leaking,
+	// when a C thread exits after a cgo call.
+	// We only invoke this function once per thread in runtime.needAndBindM,
+	// and the next calls just reuse the bound m.
+	pthread_setspecific(pthread_g, g);
+}
+
 void
 x_cgo_notify_runtime_init_done(void* dummy __attribute__ ((unused))) {
 	pthread_mutex_lock(&runtime_init_mu);
@@ -110,3 +134,14 @@ _cgo_try_pthread_create(pthread_t* thread, const pthread_attr_t* attr, void* (*p
 	}
 	return EAGAIN;
 }
+
+static void
+pthread_key_destructor(void* g) {
+	if (x_crosscall2_ptr != NULL) {
+		// fn == NULL means dropm.
+		// We restore g by using the stored g, before dropm in runtime.cgocallback,
+		// since the g stored in the TLS by Go might be cleared in some platforms,
+		// before this destructor invoked.
+		x_crosscall2_ptr(NULL, g, 0, 0);
+	}
+}
diff --git a/src/runtime/cgo/gcc_libinit_windows.c b/src/runtime/cgo/gcc_libinit_windows.c
index fdcf027424..9a8c65ea29 100644
--- a/src/runtime/cgo/gcc_libinit_windows.c
+++ b/src/runtime/cgo/gcc_libinit_windows.c
@@ -30,6 +30,9 @@ static CRITICAL_SECTION runtime_init_cs;
 static HANDLE runtime_init_wait;
 static int runtime_init_done;
 
+uintptr_t x_cgo_pthread_key_created;
+void (*x_crosscall2_ptr)(void (*fn)(void *), void *, int, size_t);
+
 // Pre-initialize the runtime synchronization objects
 void
 _cgo_preinit_init() {
@@ -91,6 +94,12 @@ _cgo_wait_runtime_init_done(void) {
 	return 0;
 }
 
+// Should not be used since x_cgo_pthread_key_created will always be zero.
+void x_cgo_bindm(void* dummy) {
+	fprintf(stderr, "unexpected cgo_bindm on Windows\n");
+	abort();
+}
+
 void
 x_cgo_notify_runtime_init_done(void* dummy) {
 	 _cgo_maybe_run_preinit();
diff --git a/src/runtime/cgo/gcc_stack_darwin.c b/src/runtime/cgo/gcc_stack_darwin.c
new file mode 100644
index 0000000000..0a9038eb3b
--- /dev/null
+++ b/src/runtime/cgo/gcc_stack_darwin.c
@@ -0,0 +1,20 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#include <pthread.h>
+#include "libcgo.h"
+
+void
+x_cgo_getstackbound(uintptr bounds[2])
+{
+	void* addr;
+	size_t size;
+	pthread_t p;
+
+	p = pthread_self();
+	addr = pthread_get_stackaddr_np(p); // high address (!)
+	size = pthread_get_stacksize_np(p);
+	bounds[0] = (uintptr)addr - size;
+	bounds[1] = (uintptr)addr;
+}
diff --git a/src/runtime/cgo/gcc_stack_unix.c b/src/runtime/cgo/gcc_stack_unix.c
new file mode 100644
index 0000000000..f3fead9c9e
--- /dev/null
+++ b/src/runtime/cgo/gcc_stack_unix.c
@@ -0,0 +1,40 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build unix && !darwin
+
+#ifndef _GNU_SOURCE // pthread_getattr_np
+#define _GNU_SOURCE
+#endif
+
+#include <pthread.h>
+#include "libcgo.h"
+
+void
+x_cgo_getstackbound(uintptr bounds[2])
+{
+	pthread_attr_t attr;
+	void *addr;
+	size_t size;
+
+#if defined(__GLIBC__) || (defined(__sun) && !defined(__illumos__))
+	// pthread_getattr_np is a GNU extension supported in glibc.
+	// Solaris is not glibc but does support pthread_getattr_np
+	// (and the fallback doesn't work...). Illumos does not.
+	pthread_getattr_np(pthread_self(), &attr);  // GNU extension
+	pthread_attr_getstack(&attr, &addr, &size); // low address
+#elif defined(__illumos__)
+	pthread_attr_init(&attr);
+	pthread_attr_get_np(pthread_self(), &attr);
+	pthread_attr_getstack(&attr, &addr, &size); // low address
+#else
+	pthread_attr_init(&attr);
+	pthread_attr_getstacksize(&attr, &size);
+	addr = __builtin_frame_address(0) + 4096 - size;
+#endif
+	pthread_attr_destroy(&attr);
+
+	bounds[0] = (uintptr)addr;
+	bounds[1] = (uintptr)addr + size;
+}
diff --git a/src/runtime/cgo/gcc_stack_windows.c b/src/runtime/cgo/gcc_stack_windows.c
new file mode 100644
index 0000000000..d798cc77d1
--- /dev/null
+++ b/src/runtime/cgo/gcc_stack_windows.c
@@ -0,0 +1,7 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+#include "libcgo.h"
+
+void x_cgo_getstackbound(uintptr bounds[2]) {} // no-op for now
diff --git a/src/runtime/cgo/libcgo.h b/src/runtime/cgo/libcgo.h
index af4960e7e9..04755f0f20 100644
--- a/src/runtime/cgo/libcgo.h
+++ b/src/runtime/cgo/libcgo.h
@@ -52,6 +52,11 @@ extern void (*_cgo_thread_start)(ThreadStart *ts);
 extern void (*_cgo_sys_thread_create)(void* (*func)(void*), void* arg);
 
 /*
+ * Indicates whether a dummy pthread per-thread variable is allocated.
+ */
+extern uintptr_t *_cgo_pthread_key_created;
+
+/*
  * Creates the new operating system thread (OS, arch dependent).
  */
 void _cgo_sys_thread_start(ThreadStart *ts);
diff --git a/src/runtime/cgocall.go b/src/runtime/cgocall.go
index 8b00f3de57..a944723882 100644
--- a/src/runtime/cgocall.go
+++ b/src/runtime/cgocall.go
@@ -236,6 +236,9 @@ func cgocallbackg(fn, frame unsafe.Pointer, ctxt uintptr) {
 	savedpc := gp.syscallpc
 	exitsyscall() // coming out of cgo call
 	gp.m.incgo = false
+	if gp.m.isextra {
+		gp.m.isExtraInC = false
+	}
 
 	osPreemptExtExit(gp.m)
 
@@ -246,6 +249,9 @@ func cgocallbackg(fn, frame unsafe.Pointer, ctxt uintptr) {
 	// This is enforced by checking incgo in the schedule function.
 
 	gp.m.incgo = true
+	if gp.m.isextra {
+		gp.m.isExtraInC = true
+	}
 
 	if gp.m != checkm {
 		throw("m changed unexpectedly in cgocallbackg")
diff --git a/src/runtime/crash_cgo_test.go b/src/runtime/crash_cgo_test.go
index c31586cce0..2a07678b52 100644
--- a/src/runtime/crash_cgo_test.go
+++ b/src/runtime/crash_cgo_test.go
@@ -833,3 +833,16 @@ func TestDestructorCallbackRace(t *testing.T) {
 		t.Errorf("expected %q, but got:\n%s", want, got)
 	}
 }
+
+func TestEnsureBindM(t *testing.T) {
+	t.Parallel()
+	switch runtime.GOOS {
+	case "windows", "plan9":
+		t.Skipf("skipping bindm test on %s", runtime.GOOS)
+	}
+	got := runTestProg(t, "testprogcgo", "EnsureBindM")
+	want := "OK\n"
+	if got != want {
+		t.Errorf("expected %q, got %v", want, got)
+	}
+}
diff --git a/src/runtime/proc.go b/src/runtime/proc.go
index fd892115bf..35aeb2d1ac 100644
--- a/src/runtime/proc.go
+++ b/src/runtime/proc.go
@@ -209,6 +209,10 @@ func main() {
 
 	main_init_done = make(chan bool)
 	if iscgo {
+		if _cgo_pthread_key_created == nil {
+			throw("_cgo_pthread_key_created missing")
+		}
+
 		if _cgo_thread_start == nil {
 			throw("_cgo_thread_start missing")
 		}
@@ -223,6 +227,13 @@ func main() {
 		if _cgo_notify_runtime_init_done == nil {
 			throw("_cgo_notify_runtime_init_done missing")
 		}
+
+		// Set the x_crosscall2_ptr C function pointer variable point to crosscall2.
+		if set_crosscall2 == nil {
+			throw("set_crosscall2 missing")
+		}
+		set_crosscall2()
+
 		// Start the template thread in case we enter Go from
 		// a C-created thread and need to create a new thread.
 		startTemplateThread()
@@ -1886,11 +1897,15 @@ func allocm(pp *p, fn func(), id int64) *m {
 // pressed into service as the scheduling stack and current
 // goroutine for the duration of the cgo callback.
 //
-// When the callback is done with the m, it calls dropm to
-// put the m back on the list.
+// It calls dropm to put the m back on the list,
+// 1. when the callback is done with the m in non-pthread platforms,
+// 2. or when the C thread exiting on pthread platforms.
+//
+// The signal argument indicates whether we're called from a signal
+// handler.
 //
 //go:nosplit
-func needm() {
+func needm(signal bool) {
 	if (iscgo || GOOS == "windows") && !cgoHasExtraM {
 		// Can happen if C/C++ code calls Go from a global ctor.
 		// Can also happen on Windows if a global ctor uses a
@@ -1936,16 +1951,36 @@ func needm() {
 	osSetupTLS(mp)
 
 	// Install g (= m->g0) and set the stack bounds
-	// to match the current stack. We don't actually know
+	// to match the current stack. If we don't actually know
 	// how big the stack is, like we don't know how big any
-	// scheduling stack is, but we assume there's at least 32 kB,
-	// which is more than enough for us.
+	// scheduling stack is, but we assume there's at least 32 kB.
+	// If we can get a more accurate stack bound from pthread,
+	// use that.
 	setg(mp.g0)
 	gp := getg()
 	gp.stack.hi = getcallersp() + 1024
 	gp.stack.lo = getcallersp() - 32*1024
+	if !signal && _cgo_getstackbound != nil {
+		// Don't adjust if called from the signal handler.
+		// We are on the signal stack, not the pthread stack.
+		// (We could get the stack bounds from sigaltstack, but
+		// we're getting out of the signal handler very soon
+		// anyway. Not worth it.)
+		var bounds [2]uintptr
+		asmcgocall(_cgo_getstackbound, unsafe.Pointer(&bounds))
+		// getstackbound is an unsupported no-op on Windows.
+		if bounds[0] != 0 {
+			gp.stack.lo = bounds[0]
+			gp.stack.hi = bounds[1]
+		}
+	}
 	gp.stackguard0 = gp.stack.lo + stackGuard
 
+	// Should mark we are already in Go now.
+	// Otherwise, we may call needm again when we get a signal, before cgocallbackg1,
+	// which means the extram list may be empty, that will cause a deadlock.
+	mp.isExtraInC = false
+
 	// Initialize this thread to use the m.
 	asminit()
 	minit()
@@ -1955,6 +1990,17 @@ func needm() {
 	sched.ngsys.Add(-1)
 }
 
+// Acquire an extra m and bind it to the C thread when a pthread key has been created.
+//
+//go:nosplit
+func needAndBindM() {
+	needm(false)
+
+	if _cgo_pthread_key_created != nil && *(*uintptr)(_cgo_pthread_key_created) != 0 {
+		cgoBindM()
+	}
+}
+
 // newextram allocates m's and puts them on the extra list.
 // It is called with a working local m, so that it can do things
 // like call schedlock and allocate.
@@ -1995,6 +2041,8 @@ func oneNewExtraM() {
 	gp.m = mp
 	mp.curg = gp
 	mp.isextra = true
+	// mark we are in C by default.
+	mp.isExtraInC = true
 	mp.lockedInt++
 	mp.lockedg.set(gp)
 	gp.lockedm.set(mp)
@@ -2018,9 +2066,11 @@ func oneNewExtraM() {
 	addExtraM(mp)
 }
 
+// dropm puts the current m back onto the extra list.
+//
+// 1. On systems without pthreads, like Windows
 // dropm is called when a cgo callback has called needm but is now
 // done with the callback and returning back into the non-Go thread.
-// It puts the current m back onto the extra list.
 //
 // The main expense here is the call to signalstack to release the
 // m's signal stack, and then the call to needm on the next callback
@@ -2032,15 +2082,18 @@ func oneNewExtraM() {
 // call. These should typically not be scheduling operations, just a few
 // atomics, so the cost should be small.
 //
-// TODO(rsc): An alternative would be to allocate a dummy pthread per-thread
-// variable using pthread_key_create. Unlike the pthread keys we already use
-// on OS X, this dummy key would never be read by Go code. It would exist
-// only so that we could register at thread-exit-time destructor.
-// That destructor would put the m back onto the extra list.
-// This is purely a performance optimization. The current version,
-// in which dropm happens on each cgo call, is still correct too.
-// We may have to keep the current version on systems with cgo
-// but without pthreads, like Windows.
+// 2. On systems with pthreads
+// dropm is called while a non-Go thread is exiting.
+// We allocate a pthread per-thread variable using pthread_key_create,
+// to register a thread-exit-time destructor.
+// And store the g into a thread-specific value associated with the pthread key,
+// when first return back to C.
+// So that the destructor would invoke dropm while the non-Go thread is exiting.
+// This is much faster since it avoids expensive signal-related syscalls.
+//
+// NOTE: this always runs without a P, so, nowritebarrierrec required.
+//
+//go:nowritebarrierrec
 func dropm() {
 	// Clear m and g, and return m to the extra list.
 	// After the call to setg we can only call nosplit functions
@@ -2067,6 +2120,39 @@ func dropm() {
 	msigrestore(sigmask)
 }
 
+// bindm store the g0 of the current m into a thread-specific value.
+//
+// We allocate a pthread per-thread variable using pthread_key_create,
+// to register a thread-exit-time destructor.
+// We are here setting the thread-specific value of the pthread key, to enable the destructor.
+// So that the pthread_key_destructor would dropm while the C thread is exiting.
+//
+// And the saved g will be used in pthread_key_destructor,
+// since the g stored in the TLS by Go might be cleared in some platforms,
+// before the destructor invoked, so, we restore g by the stored g, before dropm.
+//
+// We store g0 instead of m, to make the assembly code simpler,
+// since we need to restore g0 in runtime.cgocallback.
+//
+// On systems without pthreads, like Windows, bindm shouldn't be used.
+//
+// NOTE: this always runs without a P, so, nowritebarrierrec required.
+//
+//go:nosplit
+//go:nowritebarrierrec
+func cgoBindM() {
+	if GOOS == "windows" || GOOS == "plan9" {
+		fatal("bindm in unexpected GOOS")
+	}
+	g := getg()
+	if g.m.g0 != g {
+		fatal("the current g is not g0")
+	}
+	if _cgo_bindm != nil {
+		asmcgocall(_cgo_bindm, unsafe.Pointer(g))
+	}
+}
+
 // A helper function for EnsureDropM.
 func getm() uintptr {
 	return uintptr(unsafe.Pointer(getg().m))
diff --git a/src/runtime/runtime2.go b/src/runtime/runtime2.go
index 314ab194e7..a2075dddef 100644
--- a/src/runtime/runtime2.go
+++ b/src/runtime/runtime2.go
@@ -561,6 +561,7 @@ type m struct {
 	printlock     int8
 	incgo         bool          // m is executing a cgo call
 	isextra       bool          // m is an extra m
+	isExtraInC    bool          // m is an extra m that is not executing Go code
 	freeWait      atomic.Uint32 // Whether it is safe to free g0 and delete m (one of freeMRef, freeMStack, freeMWait)
 	fastrand      uint64
 	needextram    bool
diff --git a/src/runtime/signal_unix.go b/src/runtime/signal_unix.go
index 8a745ecda0..6ebfbbc5be 100644
--- a/src/runtime/signal_unix.go
+++ b/src/runtime/signal_unix.go
@@ -435,7 +435,7 @@ func sigtrampgo(sig uint32, info *siginfo, ctx unsafe.Pointer) {
 	c := &sigctxt{info, ctx}
 	gp := sigFetchG(c)
 	setg(gp)
-	if gp == nil {
+	if gp == nil || (gp.m != nil && gp.m.isExtraInC) {
 		if sig == _SIGPROF {
 			// Some platforms (Linux) have per-thread timers, which we use in
 			// combination with the process-wide timer. Avoid double-counting.
@@ -458,7 +458,18 @@ func sigtrampgo(sig uint32, info *siginfo, ctx unsafe.Pointer) {
 			return
 		}
 		c.fixsigcode(sig)
+		// Set g to nil here and badsignal will use g0 by needm.
+		// TODO: reuse the current m here by using the gsignal and adjustSignalStack,
+		// since the current g maybe a normal goroutine and actually running on the signal stack,
+		// it may hit stack split that is not expected here.
+		if gp != nil {
+			setg(nil)
+		}
 		badsignal(uintptr(sig), c)
+		// Restore g
+		if gp != nil {
+			setg(gp)
+		}
 		return
 	}
 
@@ -574,11 +585,11 @@ func adjustSignalStack(sig uint32, mp *m, gsigStack *gsignalStack) bool {
 
 	// sp is not within gsignal stack, g0 stack, or sigaltstack. Bad.
 	setg(nil)
-	needm()
+	needm(true)
 	if st.ss_flags&_SS_DISABLE != 0 {
 		noSignalStack(sig)
 	} else {
-		sigNotOnStack(sig)
+		sigNotOnStack(sig, sp, mp)
 	}
 	dropm()
 	return false
@@ -1018,8 +1029,10 @@ func noSignalStack(sig uint32) {
 // This is called if we receive a signal when there is a signal stack
 // but we are not on it. This can only happen if non-Go code called
 // sigaction without setting the SS_ONSTACK flag.
-func sigNotOnStack(sig uint32) {
+func sigNotOnStack(sig uint32, sp uintptr, mp *m) {
 	println("signal", sig, "received but handler not on signal stack")
+	print("mp.gsignal stack [", hex(mp.gsignal.stack.lo), " ", hex(mp.gsignal.stack.hi), "], ")
+	print("mp.g0 stack [", hex(mp.g0.stack.lo), " ", hex(mp.g0.stack.hi), "], sp=", hex(sp), "\n")
 	throw("non-Go code set up signal handler without SA_ONSTACK flag")
 }
 
@@ -1047,7 +1060,7 @@ func badsignal(sig uintptr, c *sigctxt) {
 		exit(2)
 		*(*uintptr)(unsafe.Pointer(uintptr(123))) = 2
 	}
-	needm()
+	needm(true)
 	if !sigsend(uint32(sig)) {
 		// A foreign thread received the signal sig, and the
 		// Go code does not want to handle it.
@@ -1115,8 +1128,9 @@ func sigfwdgo(sig uint32, info *siginfo, ctx unsafe.Pointer) bool {
 	//   (1) we weren't in VDSO page,
 	//   (2) we were in a goroutine (i.e., m.curg != nil), and
 	//   (3) we weren't in CGO.
+	//   (4) we weren't in dropped extra m.
 	gp := sigFetchG(c)
-	if gp != nil && gp.m != nil && gp.m.curg != nil && !gp.m.incgo {
+	if gp != nil && gp.m != nil && gp.m.curg != nil && !gp.m.isExtraInC && !gp.m.incgo {
 		return false
 	}
 
diff --git a/src/runtime/stubs.go b/src/runtime/stubs.go
index 373445d613..65b7299f74 100644
--- a/src/runtime/stubs.go
+++ b/src/runtime/stubs.go
@@ -237,6 +237,9 @@ func noEscapePtr[T any](p *T) *T {
 // cgocallback is not called from Go, only from crosscall2.
 // This in turn calls cgocallbackg, which is where we'll find
 // pointer-declared arguments.
+//
+// When fn is nil (frame is saved g), call dropm instead,
+// this is used when the C thread is exiting.
 func cgocallback(fn, frame, ctxt uintptr)
 
 func gogo(buf *gobuf)
diff --git a/src/runtime/testdata/testprogcgo/bindm.c b/src/runtime/testdata/testprogcgo/bindm.c
new file mode 100644
index 0000000000..815d8a75f2
--- /dev/null
+++ b/src/runtime/testdata/testprogcgo/bindm.c
@@ -0,0 +1,34 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build !plan9 && !windows
+
+#include <stdint.h>
+#include <pthread.h>
+#include <unistd.h>
+#include "_cgo_export.h"
+
+#define CTHREADS 2
+#define CHECKCALLS 100
+
+static void* checkBindMThread(void* thread) {
+	int i;
+	for (i = 0; i < CHECKCALLS; i++) {
+		GoCheckBindM((uintptr_t)thread);
+		usleep(1);
+	}
+	return NULL;
+}
+
+void CheckBindM() {
+	int i;
+	pthread_t s[CTHREADS];
+
+	for (i = 0; i < CTHREADS; i++) {
+		pthread_create(&s[i], NULL, checkBindMThread, &s[i]);
+	}
+	for (i = 0; i < CTHREADS; i++) {
+		pthread_join(s[i], NULL);
+	}
+}
diff --git a/src/runtime/testdata/testprogcgo/bindm.go b/src/runtime/testdata/testprogcgo/bindm.go
new file mode 100644
index 0000000000..c2003c2093
--- /dev/null
+++ b/src/runtime/testdata/testprogcgo/bindm.go
@@ -0,0 +1,61 @@
+// Copyright 2023 The Go Authors. All rights reserved.
+// Use of this source code is governed by a BSD-style
+// license that can be found in the LICENSE file.
+
+//go:build !plan9 && !windows
+
+// Test that callbacks from C to Go in the same C-thread always get the same m.
+// Make sure the extra M bind to the C-thread.
+
+package main
+
+/*
+extern void CheckBindM();
+*/
+import "C"
+
+import (
+	"fmt"
+	"os"
+	"runtime"
+	"sync"
+	"sync/atomic"
+)
+
+var (
+	mutex      = sync.Mutex{}
+	cThreadToM = map[uintptr]uintptr{}
+	started    = atomic.Uint32{}
+)
+
+// same as CTHREADS in C, make sure all the C threads are actually started.
+const cThreadNum = 2
+
+func init() {
+	register("EnsureBindM", EnsureBindM)
+}
+
+//export GoCheckBindM
+func GoCheckBindM(thread uintptr) {
+	// Wait all threads start
+	if started.Load() != cThreadNum {
+		// Only once for each thread, since it will wait all threads start.
+		started.Add(1)
+		for started.Load() < cThreadNum {
+			runtime.Gosched()
+		}
+	}
+	m := runtime_getm_for_test()
+	mutex.Lock()
+	defer mutex.Unlock()
+	if savedM, ok := cThreadToM[thread]; ok && savedM != m {
+		fmt.Printf("m == %x want %x\n", m, savedM)
+		os.Exit(1)
+	}
+	cThreadToM[thread] = m
+}
+
+func EnsureBindM() {
+	C.CheckBindM()
+	fmt.Println("OK")
+}