runtime: use fixed TLS offsets on darwin/amd64 and darwin/386

Fixes #23617

Note that this CL does not affect darwin/arm and darwin/arm64,
still TBD what, if anything, needs to be done for those.

This is a fix of CL 105975 which was reverted in CL 106155.
Needed to use movl instead of movq for 386.

Change-Id: I0db7f8087173869e60cc22c6c3124fa0a0739b46
Reviewed-on: https://go-review.googlesource.com/106156
Run-TryBot: Keith Randall <khr@golang.org>
Reviewed-by: Ian Lance Taylor <iant@golang.org>
Reviewed-by: Brad Fitzpatrick <bradfitz@golang.org>
TryBot-Result: Gobot Gobot <gobot@golang.org>

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

@@ -113,24 +113,30 @@ func (ctxt *Link) computeTLSOffset() {
 
 		/*
 		 * OS X system constants - offset from 0(GS) to our TLS.
-		 * Explained in src/runtime/cgo/gcc_darwin_*.c.
 		 */
 	case objabi.Hdarwin:
 		switch ctxt.Arch.Family {
 		default:
 			log.Fatalf("unknown thread-local storage offset for darwin/%s", ctxt.Arch.Name)
 
-		case sys.ARM:
-			ctxt.Tlsoffset = 0 // dummy value, not needed
+			/*
+			 * For x86, Apple has reserved a slot in the TLS for Go. See issue 23617.
+			 * That slot is at offset 0x30 on amd64, and 0x18 on 386.
+			 * The slot will hold the G pointer.
+			 * These constants should match those in runtime/sys_darwin_{386,amd64}.s
+			 * and runtime/cgo/gcc_darwin_{386,amd64}.c.
+			 */
+		case sys.I386:
+			ctxt.Tlsoffset = 0x18
 
 		case sys.AMD64:
-			ctxt.Tlsoffset = 0x8a0
+			ctxt.Tlsoffset = 0x30
 
-		case sys.ARM64:
+		case sys.ARM:
 			ctxt.Tlsoffset = 0 // dummy value, not needed
 
-		case sys.I386:
-			ctxt.Tlsoffset = 0x468
+		case sys.ARM64:
+			ctxt.Tlsoffset = 0 // dummy value, not needed
 		}
 	}
 

src/runtime/cgo/gcc_darwin_386.c

@@ -9,89 +9,6 @@
 #include "libcgo_unix.h"
 
 static void* threadentry(void*);
-static pthread_key_t k1;
-
-#define magic1 (0x23581321U)
-
-static void
-inittls(void)
-{
-	uint32 x;
-	pthread_key_t tofree[128], k;
-	int i, ntofree;
-
-	/*
-	 * Allocate thread-local storage slot for g.
-	 * The key numbers start at 0x100, and we expect to be
-	 * one of the early calls to pthread_key_create, so we
-	 * should be able to get a pretty low number.
-	 *
-	 * In Darwin/386 pthreads, %gs points at the thread
-	 * structure, and each key is an index into the thread-local
-	 * storage array that begins at offset 0x48 within in that structure.
-	 * It may happen that we are not quite the first function to try
-	 * to allocate thread-local storage keys, so instead of depending
-	 * on getting 0x100, we try for 0x108, allocating keys until
-	 * we get the one we want and then freeing the ones we didn't want.
-	 *
-	 * Thus the final offset to use in %gs references is
-	 * 0x48+4*0x108 = 0x468.
-	 *
-	 * The linker and runtime hard-code this constant offset
-	 * from %gs where we expect to find g.
-	 * Known to src/cmd/link/internal/ld/sym.go:/0x468
-	 * and to src/runtime/sys_darwin_386.s:/0x468
-	 *
-	 * This is truly disgusting and a bit fragile, but taking care
-	 * of it here protects the rest of the system from damage.
-	 * The alternative would be to use a global variable that
-	 * held the offset and refer to that variable each time we
-	 * need a %gs variable (g).  That approach would
-	 * require an extra instruction and memory reference in
-	 * every stack growth prolog and would also require
-	 * rewriting the code that 8c generates for extern registers.
-	 *
-	 * Things get more disgusting on OS X 10.7 Lion.
-	 * The 0x48 base mentioned above is the offset of the tsd
-	 * array within the per-thread structure on Leopard and Snow Leopard.
-	 * On Lion, the base moved a little, so while the math above
-	 * still applies, the base is different.  Thus, we cannot
-	 * look for specific key values if we want to build binaries
-	 * that run on both systems.  Instead, forget about the
-	 * specific key values and just allocate and initialize per-thread
-	 * storage until we find a key that writes to the memory location
-	 * we want.  Then keep that key.
-	 */
-	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:0x468, %0" : "=r"(x));
-		pthread_setspecific(k, 0);
-		if(x == magic1) {
-			k1 = k;
-			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;
-	}
-
-	/*
-	 * We got the key we wanted.  Free the others.
-	 */
-	for(i=0; i<ntofree; i++)
-		pthread_key_delete(tofree[i]);
-}
 
 void
 x_cgo_init(G *g)
@@ -103,8 +20,6 @@ x_cgo_init(G *g)
 	pthread_attr_getstacksize(&attr, &size);
 	g->stacklo = (uintptr)&attr - size + 4096;
 	pthread_attr_destroy(&attr);
-
-	inittls();
 }
 
 
@@ -142,10 +57,9 @@ threadentry(void *v)
 	ts = *(ThreadStart*)v;
 	free(v);
 
-	if (pthread_setspecific(k1, (void*)ts.g) != 0) {
-		fprintf(stderr, "runtime/cgo: pthread_setspecific failed\n");
-		abort();
-	}
+	// Move the g pointer into the slot reserved in thread local storage.
+	// Constant must match the one in cmd/link/internal/ld/sym.go.
+	asm volatile("movl %0, %%gs:0x18" :: "r"(ts.g));
 
 	crosscall_386(ts.fn);
 	return nil;

src/runtime/cgo/gcc_darwin_amd64.c

@@ -9,60 +9,6 @@
 #include "libcgo_unix.h"
 
 static void* threadentry(void*);
-static pthread_key_t k1;
-
-#define magic1 (0x23581321345589ULL)
-
-static void
-inittls(void)
-{
-	uint64 x;
-	pthread_key_t tofree[128], k;
-	int i, ntofree;
-
-	/*
-	 * Same logic, code as darwin_386.c:/inittls, except that words
-	 * are 8 bytes long now, and the thread-local storage starts
-	 * at 0x60 on Leopard / Snow Leopard. So the offset is
-	 * 0x60+8*0x108 = 0x8a0.
-	 *
-	 * The linker and runtime hard-code this constant offset
-	 * from %gs where we expect to find g.
-	 * Known to src/cmd/link/internal/ld/sym.go:/0x8a0
-	 * and to src/runtime/sys_darwin_amd64.s:/0x8a0
-	 *
-	 * As disgusting as on the 386; same justification.
-	 */
-	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 %%gs:0x8a0, %0" : "=r"(x));
-		pthread_setspecific(k, 0);
-		if(x == magic1) {
-			k1 = k;
-			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;
-	}
-
-	/*
-	 * We got the key we wanted.  Free the others.
-	 */
-	for(i=0; i<ntofree; i++)
-		pthread_key_delete(tofree[i]);
-}
 
 void
 x_cgo_init(G *g)
@@ -74,8 +20,6 @@ x_cgo_init(G *g)
 	pthread_attr_getstacksize(&attr, &size);
 	g->stacklo = (uintptr)&attr - size + 4096;
 	pthread_attr_destroy(&attr);
-
-	inittls();
 }
 
 
@@ -113,10 +57,9 @@ threadentry(void *v)
 	ts = *(ThreadStart*)v;
 	free(v);
 
-	if (pthread_setspecific(k1, (void*)ts.g) != 0) {
-		fprintf(stderr, "runtime/cgo: pthread_setspecific failed\n");
-		abort();
-	}
+	// Move the g pointer into the slot reserved in thread local storage.
+	// Constant must match the one in cmd/link/internal/ld/sym.go.
+	asm volatile("movq %0, %%gs:0x30" :: "r"(ts.g));
 
 	crosscall_amd64(ts.fn);
 	return nil;

src/runtime/sys_darwin_386.s

@@ -525,15 +525,14 @@ TEXT runtime·setldt(SB),NOSPLIT,$32
 	 * When linking against the system libraries,
 	 * we use its pthread_create and let it set up %gs
 	 * for us.  When we do that, the private storage
-	 * we get is not at 0(GS) but at 0x468(GS).
-	 * 8l rewrites 0(TLS) into 0x468(GS) for us.
+	 * we get is not at 0(GS) but at 0x18(GS).
+	 * The linker rewrites 0(TLS) into 0x18(GS) for us.
 	 * To accommodate that rewrite, we translate the
-	 * address and limit here so that 0x468(GS) maps to 0(address).
+	 * address here so that 0x18(GS) maps to 0(address).
 	 *
-	 * See cgo/gcc_darwin_386.c:/468 for the derivation
-	 * of the constant.
+	 * Constant must match the one in cmd/link/internal/ld/sym.go.
 	 */
-	SUBL	$0x468, BX
+	SUBL	$0x18, BX
 
 	/*
 	 * Must set up as USER_CTHREAD segment because

src/runtime/sys_darwin_amd64.s

@@ -569,11 +569,10 @@ TEXT runtime·mach_semaphore_signal_all(SB),NOSPLIT,$0
 // set tls base to DI
 TEXT runtime·settls(SB),NOSPLIT,$32
 	/*
-	* Same as in sys_darwin_386.s:/ugliness, different constant.
-	* See cgo/gcc_darwin_amd64.c for the derivation
-	* of the constant.
-	*/
-	SUBQ $0x8a0, DI
+	 * Same as in sys_darwin_386.s, but a different constant.
+	 * Constant must match the one in cmd/link/internal/ld/sym.go.
+	 */
+	SUBQ $0x30, DI
 
 	MOVL	$(0x3000000+3), AX	// thread_fast_set_cthread_self - machdep call #3
 	SYSCALL