cmd/internal/obj/riscv,cmd/link: shorten the riscv64 call sequence

Now that the other dependent offset has been identified, we can remove the
unnecessary ADDI instruction from the riscv64 call sequence (reducing it
to AUIPC+JALR, rather than the previous AUIPC+ADDI+JALR).

Change-Id: I348c4efb686f9f71ed1dd1d25fb9142a41230b0d
Reviewed-on: https://go-review.googlesource.com/c/go/+/216798
Reviewed-by: Cherry Zhang <[email protected]>
Run-TryBot: Cherry Zhang <[email protected]>
TryBot-Result: Gobot Gobot <[email protected]>

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

@@ -55,15 +55,6 @@ func jalrToSym(ctxt *obj.Link, p *obj.Prog, newprog obj.ProgAlloc, lr int16) *ob
 	p.Mark |= NEED_PCREL_ITYPE_RELOC
 	p = obj.Appendp(p, newprog)
 
-	// TODO(jsing): This instruction is not necessary, as the lower bits
-	// of the immediate can be encoded directly in the JALR instruction.
-	// However, other code currently depends on jalrToSym being 12 bytes...
-	p.As = AADDI
-	p.From = obj.Addr{Type: obj.TYPE_CONST}
-	p.Reg = REG_TMP
-	p.To = obj.Addr{Type: obj.TYPE_REG, Reg: REG_TMP}
-	p = obj.Appendp(p, newprog)
-
 	// Leave Sym only for the CALL reloc in assemble.
 	p.As = AJALR
 	p.From.Type = obj.TYPE_REG

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

@@ -271,8 +271,8 @@ func (ctxt *Link) pclntab() {
 					case sys.RISCV64:
 						// TODO(jsing): The JALR instruction is marked with
 						// R_CALLRISCV, whereas the actual reloc is currently
-						// two instructions earlier for the AUIPC+AADI sequence.
-						deferreturn -= 8
+						// one instruction earlier starting with the AUIPC.
+						deferreturn -= 4
 					case sys.S390X:
 						deferreturn -= 2
 					default:

src/runtime/asm_riscv64.s

@@ -254,12 +254,11 @@ TEXT runtime·gogo(SB), NOSPLIT, $16-8
 // func jmpdefer(fv *funcval, argp uintptr)
 // called from deferreturn
 // 1. grab stored return address from the caller's frame
-// 2. sub 12 bytes to get back to JAL deferreturn
+// 2. sub 8 bytes to get back to JAL deferreturn
 // 3. JMP to fn
-// TODO(sorear): There are shorter jump sequences.  This function will need to be updated when we use them.
 TEXT runtime·jmpdefer(SB), NOSPLIT|NOFRAME, $0-16
 	MOV	0(X2), RA
-	ADD	$-12, RA
+	ADD	$-8, RA
 
 	MOV	fv+0(FP), CTXT
 	MOV	argp+8(FP), X2