cmd/internal/obj/mips,s390x,riscv: save LR after decrementing SP

Following CL 412474, for the rest of the LR architectures. On
MIPS(32/64), S390X, and RISCV, there is no single instruction that
saves the LR and decrements the SP, so we need to insert an
instruction to save the LR after decrementing the SP.

On ARM(32) and PPC64 we already use a single instruction to save
the LR and decrement the SP.

Updates #53374.

Change-Id: I5a2e211026d95edb0e0f7d084ddb784f8077b86d
Reviewed-on: https://go-review.googlesource.com/c/go/+/413428
TryBot-Result: Gopher Robot <gobot@golang.org>
Reviewed-by: Austin Clements <austin@google.com>
Run-TryBot: Cherry Mui <cherryyz@google.com>

src/cmd/asm/internal/asm/testdata/mips64.s

@@ -21,9 +21,9 @@ label0:
 	BEQ	R1, 2(PC)
 	JMP	label0+0	// JMP 3	// 1000fffd
 	BEQ	R1, 2(PC)
-	JAL	1(PC)		// CALL 1(PC)	// 0c00000e
+	JAL	1(PC)		// CALL 1(PC)	// 0c00000f
 	BEQ	R1, 2(PC)
-	JAL	label0+0	// CALL 3	// 0c000006
+	JAL	label0+0	// CALL 3	// 0c000007
 
 //	LBRA addr
 //	{
@@ -32,11 +32,11 @@ label0:
 	BEQ	R1, 2(PC)
 	JMP	0(R1)		// JMP (R1)	// 00200008
 	BEQ	R1, 2(PC)
-	JMP	foo+0(SB)	// JMP foo(SB)	// 08000018
+	JMP	foo+0(SB)	// JMP foo(SB)	// 08000019
 	BEQ	R1, 2(PC)
 	JAL	0(R1)		// CALL (R1)	// 0020f809
 	BEQ	R1, 2(PC)
-	JAL	foo+0(SB)	// CALL foo(SB)	// 0c000020
+	JAL	foo+0(SB)	// CALL foo(SB)	// 0c000021
 
 //
 // BEQ/BNE

src/cmd/internal/obj/mips/obj0.go

@@ -343,6 +343,20 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
 				q.Spadj = +autosize
 
 				q = c.ctxt.EndUnsafePoint(q, c.newprog, -1)
+
+				// On Linux, in a cgo binary we may get a SIGSETXID signal early on
+				// before the signal stack is set, as glibc doesn't allow us to block
+				// SIGSETXID. So a signal may land on the current stack and clobber
+				// the content below the SP. We store the LR again after the SP is
+				// decremented.
+				q = obj.Appendp(q, newprog)
+				q.As = mov
+				q.Pos = p.Pos
+				q.From.Type = obj.TYPE_REG
+				q.From.Reg = REGLINK
+				q.To.Type = obj.TYPE_MEM
+				q.To.Offset = 0
+				q.To.Reg = REGSP
 			}
 
 			if c.cursym.Func().Text.From.Sym.Wrapper() && c.cursym.Func().Text.Mark&LEAF == 0 {

src/cmd/internal/obj/riscv/obj.go

@@ -410,6 +410,16 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
 		prologue.Spadj = int32(stacksize)
 
 		prologue = ctxt.EndUnsafePoint(prologue, newprog, -1)
+
+		// On Linux, in a cgo binary we may get a SIGSETXID signal early on
+		// before the signal stack is set, as glibc doesn't allow us to block
+		// SIGSETXID. So a signal may land on the current stack and clobber
+		// the content below the SP. We store the LR again after the SP is
+		// decremented.
+		prologue = obj.Appendp(prologue, newprog)
+		prologue.As = AMOV
+		prologue.From = obj.Addr{Type: obj.TYPE_REG, Reg: REG_LR}
+		prologue.To = obj.Addr{Type: obj.TYPE_MEM, Reg: REG_SP, Offset: 0}
 	}
 
 	if cursym.Func().Text.From.Sym.Wrapper() {

src/cmd/internal/obj/s390x/objz.go

@@ -358,6 +358,19 @@ func preprocess(ctxt *obj.Link, cursym *obj.LSym, newprog obj.ProgAlloc) {
 				q.Spadj = autosize
 
 				q = c.ctxt.EndUnsafePoint(q, c.newprog, -1)
+
+				// On Linux, in a cgo binary we may get a SIGSETXID signal early on
+				// before the signal stack is set, as glibc doesn't allow us to block
+				// SIGSETXID. So a signal may land on the current stack and clobber
+				// the content below the SP. We store the LR again after the SP is
+				// decremented.
+				q = obj.Appendp(q, c.newprog)
+				q.As = AMOVD
+				q.From.Type = obj.TYPE_REG
+				q.From.Reg = REG_LR
+				q.To.Type = obj.TYPE_MEM
+				q.To.Reg = REGSP
+				q.To.Offset = 0
 			} else if c.cursym.Func().Text.Mark&LEAF == 0 {
 				// A very few functions that do not return to their caller
 				// (e.g. gogo) are not identified as leaves but still have