cmd/link/ld,cmd/internal/obj,runtime: make the Android TLS offset dynamic

We're going to need a different TLS offset for Android Q, so the static
offsets used for 386 and amd64 are no longer viable on Android.

Introduce runtime·tls_g and use that for indexing into TLS storage. As
an added benefit, we can then merge the TLS setup code for all android
GOARCHs.

While we're at it, remove a bunch of android special cases no longer
needed.

Updates #29674
Updates #29249 (perhaps fixes it)

Change-Id: I77c7385aec7de8f1f6a4da7c9c79999157e39572
Reviewed-on: https://go-review.googlesource.com/c/go/+/169817
TryBot-Result: Gobot Gobot <gobot@golang.org>
Reviewed-by: Cherry Zhang <cherryyz@google.com>

src/cmd/internal/obj/x86/obj6.go

@@ -41,13 +41,8 @@ import (
 
 func CanUse1InsnTLS(ctxt *obj.Link) bool {
 	if isAndroid {
-		// For android, we use a disgusting hack that assumes
-		// the thread-local storage slot for g is allocated
-		// using pthread_key_create with a fixed offset
-		// (see src/runtime/cgo/gcc_android_amd64.c).
-		// This makes access to the TLS storage (for g) doable
-		// with 1 instruction.
-		return true
+		// Android uses a global variable for the tls offset.
+		return false
 	}
 
 	if ctxt.Arch.Family == sys.I386 {
@@ -162,6 +157,18 @@ func progedit(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) {
 		}
 	}
 
+	// Android uses a tls offset determined at runtime. Rewrite
+	//	MOVQ TLS, BX
+	// to
+	//	MOVQ runtime.tls_g(SB), BX
+	if isAndroid && (p.As == AMOVQ || p.As == AMOVL) && p.From.Type == obj.TYPE_REG && p.From.Reg == REG_TLS && p.To.Type == obj.TYPE_REG && REG_AX <= p.To.Reg && p.To.Reg <= REG_R15 {
+		p.From.Type = obj.TYPE_MEM
+		p.From.Name = obj.NAME_EXTERN
+		p.From.Reg = REG_NONE
+		p.From.Sym = ctxt.Lookup("runtime.tls_g")
+		p.From.Index = REG_NONE
+	}
+
 	// TODO: Remove.
 	if ctxt.Headtype == objabi.Hwindows && ctxt.Arch.Family == sys.AMD64 || ctxt.Headtype == objabi.Hplan9 {
 		if p.From.Scale == 1 && p.From.Index == REG_TLS {
@@ -1007,6 +1014,7 @@ func load_g_cx(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc) *obj.Prog {
 	progedit(ctxt, p, newprog)
 	for p.Link != next {
 		p = p.Link
+		progedit(ctxt, p, newprog)
 	}
 
 	if p.From.Index == REG_TLS {

src/cmd/link/internal/ld/data.go

@@ -218,9 +218,7 @@ func relocsym(ctxt *Link, s *sym.Symbol) {
 				Errorf(s, "unknown reloc to %v: %d (%s)", r.Sym.Name, r.Type, sym.RelocName(ctxt.Arch, r.Type))
 			}
 		case objabi.R_TLS_LE:
-			isAndroidX86 := objabi.GOOS == "android" && (ctxt.Arch.InFamily(sys.AMD64, sys.I386))
-
-			if ctxt.LinkMode == LinkExternal && ctxt.IsELF && !isAndroidX86 {
+			if ctxt.LinkMode == LinkExternal && ctxt.IsELF {
 				r.Done = false
 				if r.Sym == nil {
 					r.Sym = ctxt.Tlsg
@@ -243,7 +241,7 @@ func relocsym(ctxt *Link, s *sym.Symbol) {
 				// related to the fact that our own TLS storage happens
 				// to take up 8 bytes.
 				o = 8 + r.Sym.Value
-			} else if ctxt.IsELF || ctxt.HeadType == objabi.Hplan9 || ctxt.HeadType == objabi.Hdarwin || isAndroidX86 {
+			} else if ctxt.IsELF || ctxt.HeadType == objabi.Hplan9 || ctxt.HeadType == objabi.Hdarwin {
 				o = int64(ctxt.Tlsoffset) + r.Add
 			} else if ctxt.HeadType == objabi.Hwindows {
 				o = r.Add
@@ -251,9 +249,7 @@ func relocsym(ctxt *Link, s *sym.Symbol) {
 				log.Fatalf("unexpected R_TLS_LE relocation for %v", ctxt.HeadType)
 			}
 		case objabi.R_TLS_IE:
-			isAndroidX86 := objabi.GOOS == "android" && (ctxt.Arch.InFamily(sys.AMD64, sys.I386))
-
-			if ctxt.LinkMode == LinkExternal && ctxt.IsELF && !isAndroidX86 {
+			if ctxt.LinkMode == LinkExternal && ctxt.IsELF {
 				r.Done = false
 				if r.Sym == nil {
 					r.Sym = ctxt.Tlsg

src/cmd/link/internal/ld/sym.go

@@ -61,6 +61,7 @@ func linknew(arch *sys.Arch) *Link {
 }
 
 // computeTLSOffset records the thread-local storage offset.
+// Not used for Android where the TLS offset is determined at runtime.
 func (ctxt *Link) computeTLSOffset() {
 	switch ctxt.HeadType {
 	default:
@@ -80,21 +81,7 @@ func (ctxt *Link) computeTLSOffset() {
 		objabi.Hopenbsd,
 		objabi.Hdragonfly,
 		objabi.Hsolaris:
-		if objabi.GOOS == "android" {
-			switch ctxt.Arch.Family {
-			case sys.AMD64:
-				// Android/amd64 constant - offset from 0(FS) to our TLS slot.
-				// Explained in src/runtime/cgo/gcc_android_*.c
-				ctxt.Tlsoffset = 0x1d0
-			case sys.I386:
-				// Android/386 constant - offset from 0(GS) to our TLS slot.
-				ctxt.Tlsoffset = 0xf8
-			default:
 		ctxt.Tlsoffset = -1 * ctxt.Arch.PtrSize
-			}
-		} else {
-			ctxt.Tlsoffset = -1 * ctxt.Arch.PtrSize
-		}
 
 	case objabi.Hnacl:
 		switch ctxt.Arch.Family {

src/runtime/asm_386.s

@@ -171,9 +171,18 @@ nocpuinfo:
 	MOVL	_cgo_init(SB), AX
 	TESTL	AX, AX
 	JZ	needtls
+#ifdef GOOS_android
+	MOVL	0(TLS), BX
+	MOVL	BX, 12(SP)	// arg 4: TLS base, stored in the first slot (TLS_SLOT_SELF).
+	MOVL	$runtime·tls_g(SB), 8(SP)	// arg 3: &tls_g
+#else
+	MOVL	$0, BX
+	MOVL	BX, 12(SP)	// arg 3,4: not used when using platform's TLS
+	MOVL	BX, 8(SP)
+#endif
 	MOVL	$setg_gcc<>(SB), BX
-	MOVL	BX, 4(SP)
-	MOVL	BP, 0(SP)
+	MOVL	BX, 4(SP)	// arg 2: setg_gcc
+	MOVL	BP, 0(SP)	// arg 1: g0
 	CALL	AX
 
 	// update stackguard after _cgo_init
@@ -1553,3 +1562,7 @@ TEXT runtime·panicExtendSlice3CU(SB),NOSPLIT,$0-12
 	MOVL	AX, lo+4(FP)
 	MOVL	CX, y+8(FP)
 	JMP	runtime·goPanicExtendSlice3CU(SB)
+
+#ifdef GOOS_android
+GLOBL runtime·tls_g+0(SB), NOPTR, $4
+#endif

src/runtime/asm_amd64.s

@@ -132,9 +132,22 @@ nocpuinfo:
 	MOVQ	_cgo_init(SB), AX
 	TESTQ	AX, AX
 	JZ	needtls
-	// g0 already in DI
-	MOVQ	DI, CX	// Win64 uses CX for first parameter
-	MOVQ	$setg_gcc<>(SB), SI
+	// arg 1: g0, already in DI
+	MOVQ	$setg_gcc<>(SB), SI // arg 2: setg_gcc
+#ifdef GOOS_android
+	MOVQ	$runtime·tls_g(SB), DX 	// arg 3: &tls_g
+	MOVQ	0(TLS), CX	// arg 4: TLS base, stored in the first slot (TLS_SLOT_SELF).
+#else
+	MOVQ	$0, DX	// arg 3, 4: not used when using platform's TLS
+	MOVQ	$0, CX
+#endif
+#ifdef GOOS_windows
+	// Adjust for the Win64 calling convention.
+	MOVQ	CX, R9 // arg 4
+	MOVQ	DX, R8 // arg 3
+	MOVQ	SI, DX // arg 2
+	MOVQ	DI, CX // arg 1
+#endif
 	CALL	AX
 
 	// update stackguard after _cgo_init
@@ -1698,3 +1711,7 @@ TEXT runtime·panicSlice3CU(SB),NOSPLIT,$0-16
 	MOVQ	AX, x+0(FP)
 	MOVQ	CX, y+8(FP)
 	JMP	runtime·goPanicSlice3CU(SB)
+
+#ifdef GOOS_android
+GLOBL runtime·tls_g+0(SB), NOPTR, $8
+#endif

src/runtime/cgo/gcc_android.c

@@ -4,6 +4,7 @@
 
 #include <stdarg.h>
 #include <android/log.h>
+#include <pthread.h>
 #include "libcgo.h"
 
 void
@@ -29,3 +30,37 @@ fatalf(const char* format, ...)
 
 	abort();
 }
+
+// Truncated to a different magic value on 32-bit; that's ok.
+#define magic1 (0x23581321345589ULL)
+
+// inittls allocates a thread-local storage slot for g.
+//
+// It finds the first available slot using pthread_key_create and uses
+// it as the offset value for runtime.tls_g.
+static void
+inittls(void **tlsg, void **tlsbase)
+{
+	pthread_key_t k;
+	int i, err;
+
+	err = pthread_key_create(&k, nil);
+	if(err != 0) {
+		fatalf("pthread_key_create failed: %d", err);
+	}
+	pthread_setspecific(k, (void*)magic1);
+	// If thread local slots are laid out as we expect, our magic word will
+	// be located at some low offset from tlsbase. However, just in case something went
+	// wrong, the search is limited to sensible offsets. PTHREAD_KEYS_MAX was the
+	// original limit, but issue 19472 made a higher limit necessary.
+	for (i=0; i<384; i++) {
+		if (*(tlsbase+i) == (void*)magic1) {
+			*tlsg = (void*)(i*sizeof(void *));
+			pthread_setspecific(k, 0);
+			return;
+		}
+	}
+	fatalf("could not find pthread key");
+}
+
+void (*x_cgo_inittls)(void **tlsg, void **tlsbase) = inittls;

src/runtime/cgo/gcc_android_386.c

-// Copyright 2015 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 <string.h> /* for strerror */
-#include <pthread.h>
-#include <signal.h>
-#include "libcgo.h"
-
-#define magic1 (0x23581321U)
-
-static void
-inittls(void)
-{
-	uint32 x;
-	pthread_key_t tofree[128], k;
-	int i, ntofree;
-
-	/*
-	 * Same logic, code as gcc_android_amd64.c:/inittls.
-	 * Note that this is a temporary hack that should be fixed soon.
-	 *
-	 * TODO: fix this.
-	 *
-	 * The linker and runtime hard-code this constant offset
-	 * from %gs where we expect to find g. Disgusting.
-	 *
-	 * Known to src/cmd/link/internal/ld/sym.go:/0xf8
-	 * and to src/runtime/sys_linux_386.s:/0xf8 or /GOOS_android.
-	 * TODO(hyangah): check 0xb0 works with API23+
-	 *
-	 * As disgusting as on the darwin/386, darwin/amd64.
-	 */
-	ntofree = 0;
-	for(;;) {
-		if(pthread_key_create(&k, nil) != 0) {
-			fprintf(stderr, "runtime/cgo: pthread_key_create failed\n");
-			abort();
-		}
-		pthread_setspecific(k, (void*)magic1);
-		asm volatile("movl %%gs:0xf8, %0" : "=r"(x));
-		pthread_setspecific(k, 0);
-		if (x == magic1) {
-			break;
-		}
-		if(ntofree >= nelem(tofree)) {
-			fprintf(stderr, "runtime/cgo: could not obtain pthread_keys\n");
-			fprintf(stderr, "\ttried");
-			for(i=0; i<ntofree; i++)
-				fprintf(stderr, " %#x", (unsigned)tofree[i]);
-			fprintf(stderr, "\n");
-			abort();
-		}
-		tofree[ntofree++] = k;
-	}
-	// TODO: output to stderr is not useful for apps.
-	// Can we fall back to Android's log library?
-
-	/*
-	 * We got the key we wanted.  Free the others.
-	 */
-	for(i=0; i<ntofree; i++) {
-		pthread_key_delete(tofree[i]);
-	}
-}
-
-void (*x_cgo_inittls)(void) = inittls;

src/runtime/cgo/gcc_android_amd64.c

-// Copyright 2015 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 <string.h> /* for strerror */
-#include <pthread.h>
-#include <signal.h>
-#include "libcgo.h"
-
-#define magic1 (0x23581321345589ULL)
-
-static void
-inittls(void)
-{
-	uint64 x;
-	pthread_key_t tofree[128], k;
-	int i, ntofree;
-
-	/*
-	 * Same logic, code as gcc_darwin_386.c:/inittls.
-	 * Note that this is a temporary hack that should be fixed soon.
-	 * Android-L and M bionic's pthread implementation differ
-	 * significantly, and can change any time.
-	 * https://android-review.googlesource.com/#/c/134202
-	 *
-	 * We chose %fs:0x1d0 which seems to work in testing with Android
-	 * emulators (API22, API23) but it may break any time.
-	 *
-	 * TODO: fix this.
-	 *
-	 * The linker and runtime hard-code this constant offset
-	 * from %fs where we expect to find g. Disgusting.
-	 *
-	 * Known to src/cmd/link/internal/ld/sym.go:/0x1d0
-	 * and to src/runtime/sys_linux_amd64.s:/0x1d0 or /GOOS_android.
-	 *
-	 * As disgusting as on the darwin/386, darwin/amd64.
-	 */
-	ntofree = 0;
-	for(;;) {
-		if(pthread_key_create(&k, nil) != 0) {
-			fprintf(stderr, "runtime/cgo: pthread_key_create failed\n");
-			abort();
-		}
-		pthread_setspecific(k, (void*)magic1);
-		asm volatile("movq %%fs:0x1d0, %0" : "=r"(x));
-		pthread_setspecific(k, 0);
-		if(x == magic1) {
-			break;
-		}
-		if(ntofree >= nelem(tofree)) {
-			fprintf(stderr, "runtime/cgo: could not obtain pthread_keys\n");
-			fprintf(stderr, "\ttried");
-			for(i=0; i<ntofree; i++)
-				fprintf(stderr, " %#x", (unsigned)tofree[i]);
-			fprintf(stderr, "\n");
-			abort();
-		}
-		tofree[ntofree++] = k;
-	}
-	// TODO: output to stderr is not useful for apps.
-	// Can we fall back to Android's log library?
-
-	/*
-	 * We got the key we wanted.  Free the others.
-	 */
-	for(i=0; i<ntofree; i++) {
-		pthread_key_delete(tofree[i]);
-	}
-}
-
-void (*x_cgo_inittls)(void) = inittls;

src/runtime/cgo/gcc_android_arm.c

-// Copyright 2014 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 <signal.h>
-#include <stdio.h>
-#include <sys/limits.h>
-#include "libcgo.h"
-
-#define magic1 (0x23581321U)
-
-// inittls allocates a thread-local storage slot for g.
-//
-// It finds the first available slot using pthread_key_create and uses
-// it as the offset value for runtime.tlsg.
-static void
-inittls(void **tlsg, void **tlsbase)
-{
-	pthread_key_t k;
-	int i, err;
-
-	err = pthread_key_create(&k, nil);
-	if(err != 0) {
-		fatalf("pthread_key_create failed: %d", err);
-	}
-	pthread_setspecific(k, (void*)magic1);
-	// If thread local slots are laid out as we expect, our magic word will
-	// be located at some low offset from tlsbase. However, just in case something went
-	// wrong, the search is limited to sensible offsets. PTHREAD_KEYS_MAX was the
-	// original limit, but issue 19472 made a higher limit necessary.
-	for (i=0; i<384; i++) {
-		if (*(tlsbase+i) == (void*)magic1) {
-			*tlsg = (void*)(i*sizeof(void *));
-			pthread_setspecific(k, 0);
-			return;
-		}
-	}
-	fatalf("could not find pthread key");
-}
-
-void (*x_cgo_inittls)(void **tlsg, void **tlsbase) = inittls;

src/runtime/cgo/gcc_android_arm64.c

-// Copyright 2015 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 <signal.h>
-#include <stdio.h>
-#include <sys/limits.h>
-#include "libcgo.h"
-
-#define magic1 (0x23581321345589ULL)
-
-// inittls allocates a thread-local storage slot for g.
-//
-// It finds the first available slot using pthread_key_create and uses
-// it as the offset value for runtime.tlsg.
-static void
-inittls(void **tlsg, void **tlsbase)
-{
-	pthread_key_t k;
-	int i, err;
-
-	err = pthread_key_create(&k, nil);
-	if(err != 0) {
-		fatalf("pthread_key_create failed: %d", err);
-	}
-	pthread_setspecific(k, (void*)magic1);
-	// If thread local slots are laid out as we expect, our magic word will
-	// be located at some low offset from tlsbase. However, just in case something went
-	// wrong, the search is limited to sensible offsets. PTHREAD_KEYS_MAX was the
-	// original limit, but issue 19472 made a higher limit necessary.
-	for (i=0; i<384; i++) {
-		if (*(tlsbase+i) == (void*)magic1) {
-			*tlsg = (void*)(i*sizeof(void *));
-			pthread_setspecific(k, 0);
-			return;
-		}
-	}
-	fatalf("could not find pthread key");
-}
-
-void (*x_cgo_inittls)(void **tlsg, void **tlsbase) = inittls;

src/runtime/cgo/gcc_linux_386.c

@@ -11,11 +11,11 @@
 static void *threadentry(void*);
 static void (*setg_gcc)(void*);
 
-// This will be set in gcc_android_386.c for android-specific customization.
-void (*x_cgo_inittls)(void);
+// This will be set in gcc_android.c for android-specific customization.
+void (*x_cgo_inittls)(void **tlsg, void **tlsbase);
 
 void
-x_cgo_init(G *g, void (*setg)(void*))
+x_cgo_init(G *g, void (*setg)(void*), void **tlsg, void **tlsbase)
 {
 	pthread_attr_t attr;
 	size_t size;
@@ -27,7 +27,7 @@ x_cgo_init(G *g, void (*setg)(void*))
 	pthread_attr_destroy(&attr);
 
 	if (x_cgo_inittls) {
-		x_cgo_inittls();
+		x_cgo_inittls(tlsg, tlsbase);
 	}
 }
 

src/runtime/cgo/gcc_linux_amd64.c

@@ -13,11 +13,11 @@
 static void* threadentry(void*);
 static void (*setg_gcc)(void*);
 
-// This will be set in gcc_android_amd64.c for android-specific customization.
-void (*x_cgo_inittls)(void);
+// This will be set in gcc_android.c for android-specific customization.
+void (*x_cgo_inittls)(void **tlsg, void **tlsbase);
 
 void
-x_cgo_init(G* g, void (*setg)(void*))
+x_cgo_init(G *g, void (*setg)(void*), void **tlsg, void **tlsbase)
 {
 	pthread_attr_t *attr;
 	size_t size;
@@ -49,7 +49,7 @@ x_cgo_init(G* g, void (*setg)(void*))
 	free(attr);
 
 	if (x_cgo_inittls) {
-		x_cgo_inittls();
+		x_cgo_inittls(tlsg, tlsbase);
 	}
 }
 

src/runtime/sys_linux_386.s

@@ -575,12 +575,8 @@ TEXT runtime·setldt(SB),NOSPLIT,$32
 	MOVL	address+4(FP), DX	// base address
 
 #ifdef GOOS_android
-	/*
-	 * Same as in sys_darwin_386.s:/ugliness, different constant.
-	 * address currently holds m->tls, which must be %gs:0xf8.
-	 * See cgo/gcc_android_386.c for the derivation of the constant.
-	 */
-	SUBL	$0xf8, DX
+	// Android stores the TLS offset in runtime·tls_g.
+	SUBL	runtime·tls_g(SB), DX
 	MOVL	DX, 0(DX)
 #else
 	/*

src/runtime/sys_linux_amd64.s

@@ -605,10 +605,8 @@ TEXT runtime·sigaltstack(SB),NOSPLIT,$-8
 // set tls base to DI
 TEXT runtime·settls(SB),NOSPLIT,$32
 #ifdef GOOS_android
-	// Same as in sys_darwin_386.s:/ugliness, different constant.
-	// DI currently holds m->tls, which must be fs:0x1d0.
-	// See cgo/gcc_android_amd64.c for the derivation of the constant.
-	SUBQ	$0x1d0, DI  // In android, the tls base
+	// Android stores the TLS offset in runtime·tls_g.
+	SUBQ	runtime·tls_g(SB), DI
 #else
 	ADDQ	$8, DI	// ELF wants to use -8(FS)
 #endif