[dev.power64] cmd/ld, runtime: detect, fix nosplit overflows

LGTM=minux
R=minux
CC=golang-codereviews
https://golang.org/cl/121690043

src/cmd/ld/lib.c

@@ -1026,7 +1026,7 @@ static LSym *newstack;
 
 enum
 {
-	HasLinkRegister = (thechar == '5'),
+	HasLinkRegister = (thechar == '5' || thechar == '9'),
 };
 
 // TODO: Record enough information in new object files to
@@ -1035,7 +1035,7 @@ enum
 static int
 callsize(void)
 {
-	if(thechar == '5')
+	if(HasLinkRegister)
 		return 0;
 	return RegSize;
 }
@@ -1046,9 +1046,6 @@ dostkcheck(void)
 	Chain ch;
 	LSym *s;
 
-	if(thechar == '9')
-		return;
-	
 	morestack = linklookup(ctxt, "runtime.morestack", 0);
 	newstack = linklookup(ctxt, "runtime.newstack", 0);
 
@@ -1072,19 +1069,19 @@ dostkcheck(void)
 			continue;
 
 		if(s->nosplit) {
-		ctxt->cursym = s;
-		ch.sym = s;
-		stkcheck(&ch, 0);
-	}
+			ctxt->cursym = s;
+			ch.sym = s;
+			stkcheck(&ch, 0);
+		}
 	}
 	for(s = ctxt->textp; s != nil; s = s->next) {
 		if(!s->nosplit) {
-		ctxt->cursym = s;
-		ch.sym = s;
-		stkcheck(&ch, 0);
+			ctxt->cursym = s;
+			ch.sym = s;
+			stkcheck(&ch, 0);
+		}
 	}
 }
-}
 
 static int
 stkcheck(Chain *up, int depth)
@@ -1102,7 +1099,7 @@ stkcheck(Chain *up, int depth)
 	// function at top of safe zone once.
 	if(limit == StackLimit-callsize()) {
 		if(s->stkcheck)
-		return 0;
+			return 0;
 		s->stkcheck = 1;
 	}
 	
@@ -1161,8 +1158,8 @@ stkcheck(Chain *up, int depth)
 				// to StackLimit beyond the frame size.
 				if(strncmp(r->sym->name, "runtime.morestack", 17) == 0) {
 					limit = StackLimit + s->locals;
-					if(thechar == '5')
-						limit += 4; // saved LR
+					if(HasLinkRegister)
+						limit += RegSize;
 				}
 				break;
 
@@ -1181,7 +1178,7 @@ stkcheck(Chain *up, int depth)
 				break;
 			}
 		}
-		}
+	}
 		
 	return 0;
 }
@@ -1210,7 +1207,7 @@ stkprint(Chain *ch, int limit)
 		else
 			print("\t%d\tguaranteed after split check in %s\n", ch->limit, name);
 	} else {
-		stkprint(ch->up, ch->limit + (!HasLinkRegister)*PtrSize);
+		stkprint(ch->up, ch->limit + (!HasLinkRegister)*RegSize);
 		if(!HasLinkRegister)
 			print("\t%d\ton entry to %s\n", ch->limit, name);
 	}

src/pkg/runtime/os_linux.c

@@ -49,9 +49,22 @@ runtime·futexsleep(uint32 *addr, uint32 val, int64 ns)
 		runtime·futex(addr, FUTEX_WAIT, val, nil, nil, 0);
 		return;
 	}
-	// NOTE: tv_nsec is int64 on amd64, so this assumes a little-endian system.
+
+	// It's difficult to live within the no-split stack limits here.
+	// On ARM and 386, a 64-bit divide invokes a general software routine
+	// that needs more stack than we can afford. So we use timediv instead.
+	// But on real 64-bit systems, where words are larger but the stack limit
+	// is not, even timediv is too heavy, and we really need to use just an
+	// ordinary machine instruction.
+	// Sorry for the #ifdef.
+	// For what it's worth, the #ifdef eliminated an implicit little-endian assumption.
+#ifdef _64BIT
+	ts.tv_sec = ns / 1000000000LL;
+	ts.tv_nsec = ns % 1000000000LL;
+#else
 	ts.tv_nsec = 0;
 	ts.tv_sec = runtime·timediv(ns, 1000000000LL, (int32*)&ts.tv_nsec);
+#endif
 	runtime·futex(addr, FUTEX_WAIT, val, &ts, nil, 0);
 }
 

src/pkg/runtime/string.go

@@ -39,22 +39,18 @@ func concatstrings(a []string) string {
 	return s
 }
 
-//go:nosplit
 func concatstring2(a [2]string) string {
 	return concatstrings(a[:])
 }
 
-//go:nosplit
 func concatstring3(a [3]string) string {
 	return concatstrings(a[:])
 }
 
-//go:nosplit
 func concatstring4(a [4]string) string {
 	return concatstrings(a[:])
 }
 
-//go:nosplit
 func concatstring5(a [5]string) string {
 	return concatstrings(a[:])
 }

test/nosplit.go

@@ -126,8 +126,9 @@ main 136 nosplit; REJECT
 # Calling a nosplit function from a nosplit function requires
 # having room for the saved caller PC and the called frame.
 # Because ARM doesn't save LR in the leaf, it gets an extra 4 bytes.
+# Because Power64 doesn't save LR in the leaf, it gets an extra 8 bytes.
 main 112 nosplit call f; f 0 nosplit
-main 116 nosplit call f; f 0 nosplit; REJECT amd64
+main 116 nosplit call f; f 0 nosplit
 main 120 nosplit call f; f 0 nosplit; REJECT amd64
 main 124 nosplit call f; f 0 nosplit; REJECT amd64 386
 main 128 nosplit call f; f 0 nosplit; REJECT
@@ -136,8 +137,8 @@ main 136 nosplit call f; f 0 nosplit; REJECT
 
 # Calling a splitting function from a nosplit function requires
 # having room for the saved caller PC of the call but also the
-# saved caller PC for the call to morestack. Again the ARM works
-# in less space.
+# saved caller PC for the call to morestack.
+# Again the ARM and Power64 work in less space.
 main 104 nosplit call f; f 0 call f
 main 108 nosplit call f; f 0 call f
 main 112 nosplit call f; f 0 call f; REJECT amd64
@@ -235,7 +236,7 @@ TestCases:
 		switch goarch {
 		case "power64", "power64le":
 			ptrSize = 8
-			fmt.Fprintf(&buf, "#define CALL BL\n#define REGISTER (R0)\n#define RET RETURN\n")
+			fmt.Fprintf(&buf, "#define CALL BL\n#define REGISTER (CTR)\n#define RET RETURN\n")
 		case "arm":
 			fmt.Fprintf(&buf, "#define CALL BL\n#define REGISTER (R0)\n")
 		case "amd64":