lib: Use proper add opcode on RV32 with Zalrsc

The addw opcode is only defined in RV64, which produces 32-bit results.
On RV32, the default add opcode already produces 32-bit results.

Fixes: 995f226f3f ("lib: Emit lr and sc instructions based on -march flags")
Signed-off-by: Marti Alonso <martialonso11@gmail.com>
Reviewed-by: Anup Patel <anup@brainfault.org>
Link: https://lore.kernel.org/r/20260301205421.2074835-1-martialonso11@gmail.com
Signed-off-by: Anup Patel <anup@brainfault.org>

firmware/payloads/test_head.S

@@ -22,6 +22,7 @@
 
 #define REG_L		__REG_SEL(ld, lw)
 #define REG_S		__REG_SEL(sd, sw)
+#define REG_ADDW	__REG_SEL(addw, add)
 
 	.section .entry, "ax", %progbits
 	.align 3
@@ -35,7 +36,7 @@ _start:
 #elif defined(__riscv_zalrsc)
 _sc_fail:
 	lr.w	t0, (a3)
-	addw	t1, t0, a2
+	REG_ADDW t1, t0, a2
 	sc.w	t1, t1, (a3)
 	bnez	t1, _sc_fail
 	move	a3, t0

lib/sbi/riscv_atomic.c

@@ -48,7 +48,7 @@ long atomic_add_return(atomic_t *atom, long value)
 	long ret, temp;
 #if __SIZEOF_LONG__ == 4
 	__asm__ __volatile__("1:lr.w.aqrl	%1,%0\n"
-			     "  addw	%2,%1,%3\n"
+			     "  add	%2,%1,%3\n"
 			     "  sc.w.aqrl	%2,%2,%0\n"
 			     "  bnez	%2,1b"
 			     : "+A"(atom->counter), "=&r"(ret), "=&r"(temp)

lib/sbi/riscv_locks.c

@@ -57,7 +57,11 @@ void spin_lock(spinlock_t *lock)
 		"	amoadd.w.aqrl	%0, %4, %3\n"
 #elif defined(__riscv_zalrsc)
 		"3:	lr.w.aqrl	%0, %3\n"
+#if __riscv_xlen == 64
 		"	addw	%1, %0, %4\n"
+#elif __riscv_xlen == 32
+		"	add	%1, %0, %4\n"
+#endif
 		"	sc.w.aqrl	%1, %1, %3\n"
 		"	bnez	%1, 3b\n"
 #else