On 2026/1/23 16:07 bot+bpf-ci@kernel.org write:

quoted

diff --git a/include/linux/bpf.h b/include/linux/bpf.h
index 41228b0add52..29eecd79352e 100644
--- a/include/linux/bpf.h
+++ b/include/linux/bpf.h

@@ -1229,6 +1229,8 @@ enum {
 #endif
 };

+#define BPF_TRAMP_IS_RETURN_SHIFT	63
+
 struct bpf_tramp_links {

quoted

diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
index 19fd16697262..2ae8f29b854f 100644
--- a/kernel/bpf/verifier.c
+++ b/kernel/bpf/verifier.c

@@ -23011,6 +23011,19 @@ static int fixup_kfunc_call(struct bpf_verifier_env *env, struct bpf_insn *insn,
 		   desc->func_id == special_kfunc_list[KF_bpf_rdonly_cast]) {
 		insn_buf[0] = BPF_MOV64_REG(BPF_REG_0, BPF_REG_1);
 		*cnt = 1;
+	} else if (desc->func_id == special_kfunc_list[KF_bpf_session_is_return] &&
+		   env->prog->expected_attach_type == BPF_TRACE_FSESSION) {
+		/*
+		 * inline the bpf_session_is_return() for fsession:
+		 *   bool bpf_session_is_return(void *ctx)
+		 *   {
+		 *       return (((u64 *)ctx)[-1] >> BPF_TRAMP_IS_RETURN_SHIFT) & 1;
+		 *   }
+		 */
+		insn_buf[0] = BPF_LDX_MEM(BPF_DW, BPF_REG_0, BPF_REG_1, -8);
+		insn_buf[1] = BPF_ALU64_IMM(BPF_RSH, BPF_REG_0, BPF_TRAMP_IS_RETURN_SHIFT);
+		insn_buf[2] = BPF_ALU64_IMM(BPF_AND, BPF_REG_0, 1);
+		*cnt = 3;
+	}

The inline implementation reads bit 63 of ctx[-1] to check the is_return
flag. However, looking at the x86 trampoline code in
arch/x86/net/bpf_jit_comp.c, the trampoline only stores nr_regs in ctx[-1]:

    emit_mov_imm64(&prog, BPF_REG_0, 0, (u32) nr_regs);
    emit_stx(&prog, BPF_DW, BPF_REG_FP, BPF_REG_0, -nregs_off);

The is_return flag (bit 63) is never set by the current trampoline code.
This means bpf_session_is_return() will always return 0 for fsession
programs, even when executing at the return point.

I see this is fixed later in the series by commit ba766b26e776 ("bpf,x86:
add fsession support for x86_64") which adds:

    func_meta |= (1ULL << BPF_TRAMP_IS_RETURN_SHIFT);
    if (bpf_fsession_cnt(tlinks))
        emit_store_stack_imm64(&prog, BPF_REG_0, -func_meta_off, func_meta);

Should this commit be squashed with the x86 trampoline changes to ensure
the feature works correctly at each point in the series?

As I said before, this patch implement the common part of the fsession.
and we implement the architecture specific code for x86_64 in the later
patch.

[ ... ]


---
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/21278745581