> diff --git a/arch/riscv/net/bpf_jit_comp64.c b/arch/riscv/net/bpf_jit_comp64.c
> index c03c1de16b79..eafac0655dab 100644
> --- a/arch/riscv/net/bpf_jit_comp64.c
> +++ b/arch/riscv/net/bpf_jit_comp64.c
[ ... ]
> @@ -2002,11 +2043,61 @@ int bpf_jit_emit_insn(const struct bpf_insn *insn,
> struct rv_jit_context *ctx,
> void bpf_jit_build_prologue(struct rv_jit_context *ctx, bool is_subprog)
> {
> int i, stack_adjust = 0, store_offset, bpf_stack_adjust;
> + struct bpf_prog_aux *aux = ctx->prog->aux;
>
> bpf_stack_adjust = round_up(ctx->prog->aux->stack_depth, STACK_ALIGN);
> if (bpf_stack_adjust)
> mark_fp(ctx);
>
> + if (aux->exception_boundary || aux->exception_cb) {
> + /*
> + * A program acting as an exception boundary saves the full set
> + * of riscv callee saved registers (ra, s0-s11).
> + */
> + stack_adjust = round_up(ARRAY_SIZE(rv_exception_csave_regs) * 8,
> + STACK_ALIGN);
> + stack_adjust += bpf_stack_adjust;
> + store_offset = stack_adjust - 8;
> +
> + /* emit kcfi type preamble immediately before the first insn */
> + emit_kcfi(is_subprog ? cfi_bpf_subprog_hash : cfi_bpf_hash,
> ctx);
> +
> + /* nops reserved for auipc+jalr pair */
> + for (i = 0; i < RV_FENTRY_NINSNS; i++)
> + emit(rv_nop(), ctx);
> +
> + /* First instruction is always setting the tail-call-counter
> + * (TCC) register. This instruction is skipped for tail calls.
> + * Force using a 4-byte (non-compressed) instruction.
> + */
> + emit(rv_addi(RV_REG_TCC, RV_REG_ZERO, MAX_TAIL_CALL_CNT), ctx);
This isn't a bug, but could the kcfi preamble, fentry-nops loop, and
TCC-init instruction be emitted once before the exception_boundary /
exception_cb branch instead of being duplicated? The same three-part
sequence appears again below in the non-exception path, and the two copies
must stay byte-for-byte identical since the tail-call epilogue skip offset
depends on this fixed layout.
> +
> + if (!aux->exception_cb && aux->exception_boundary) {
> + /*
> + * Boundary program: allocate the frame and save the
> + * full callee-saved set, capturing the caller's values.
> + */
> + emit_addi(RV_REG_SP, RV_REG_SP, -stack_adjust, ctx);
> + for (i = 0; i < ARRAY_SIZE(rv_exception_csave_regs);
> i++) {
> + emit_sd(RV_REG_SP, store_offset,
> + rv_exception_csave_regs[i], ctx);
> + store_offset -= 8;
> + }
> + emit_addi(RV_REG_FP, RV_REG_SP, stack_adjust, ctx);
> + } else {
> + /*
> + * Exception callback, reuse the boundary program's
> + * frame, whose frame pointer is passed in a2. Setting
> + * SP = FP - stack_adjust lines the epilogue's loads up
> + * with the registers the boundary saved.
> + */
> + emit_mv(RV_REG_FP, RV_REG_A2, ctx);
> + emit_addi(RV_REG_SP, RV_REG_FP, -stack_adjust, ctx);
> + }
> +
> + goto tail_setup;
> + }
> +
> if (seen_reg(RV_REG_RA, ctx))
> stack_adjust += 8;
> stack_adjust += 8; /* RV_REG_FP */
[ ... ]
---
AI reviewed your patch. Please fix the bug or email reply why it's not a bug.
See: https://github.com/kernel-patches/vmtest/blob/master/ci/claude/README.md
CI run summary: https://github.com/kernel-patches/bpf/actions/runs/27908252575
