riscv, bpf: Track both a0 (RISC-V ABI) and a5 (BPF) return values

The RISC-V BPF uses a5 for BPF return values, which are zero-extended,
whereas the RISC-V ABI uses a0 which is sign-extended. In other words,
a5 and a0 can differ, and are used in different context.

The BPF trampoline are used for both BPF programs, and regular kernel
functions.

Make sure that the RISC-V BPF trampoline saves, and restores both a0
and a5.

Fixes: 49b5e77 ("riscv, bpf: Add bpf trampoline support for RV64")
Signed-off-by: Björn Töpel <bjorn@rivosinc.com>
Signed-off-by: Daniel Borkmann <daniel@iogearbox.net>
Link: https://lore.kernel.org/bpf/20231004120706.52848-3-bjorn@kernel.org

Author: bjorn-rivos

arch/riscv/net/bpf_jit_comp64.c

@@ -759,8 +759,10 @@ static int invoke_bpf_prog(struct bpf_tramp_link *l, int args_off, int retval_of
 	if (ret)
 		return ret;
 
-	if (save_ret)
-		emit_sd(RV_REG_FP, -retval_off, regmap[BPF_REG_0], ctx);
+	if (save_ret) {
+		emit_sd(RV_REG_FP, -retval_off, RV_REG_A0, ctx);
+		emit_sd(RV_REG_FP, -(retval_off - 8), regmap[BPF_REG_0], ctx);
+	}
 
 	/* update branch with beqz */
 	if (ctx->insns) {
@@ -853,7 +855,7 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
 
 	save_ret = flags & (BPF_TRAMP_F_CALL_ORIG | BPF_TRAMP_F_RET_FENTRY_RET);
 	if (save_ret) {
-		stack_size += 8;
+		stack_size += 16; /* Save both A5 (BPF R0) and A0 */
 		retval_off = stack_size;
 	}
 
@@ -957,6 +959,7 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
 		if (ret)
 			goto out;
 		emit_sd(RV_REG_FP, -retval_off, RV_REG_A0, ctx);
+		emit_sd(RV_REG_FP, -(retval_off - 8), regmap[BPF_REG_0], ctx);
 		im->ip_after_call = ctx->insns + ctx->ninsns;
 		/* 2 nops reserved for auipc+jalr pair */
 		emit(rv_nop(), ctx);
@@ -988,8 +991,10 @@ static int __arch_prepare_bpf_trampoline(struct bpf_tramp_image *im,
 	if (flags & BPF_TRAMP_F_RESTORE_REGS)
 		restore_args(nregs, args_off, ctx);
 
-	if (save_ret)
+	if (save_ret) {
 		emit_ld(RV_REG_A0, -retval_off, RV_REG_FP, ctx);
+		emit_ld(regmap[BPF_REG_0], -(retval_off - 8), RV_REG_FP, ctx);
+	}
 
 	emit_ld(RV_REG_S1, -sreg_off, RV_REG_FP, ctx);