Re: [PATCH bpf-next 1/1] bpf: Allow fall back to interpreter for programs with stack size <= 512

[Yonghong Song <yonghong.song@linux.dev>]

On 8/5/25 4:55 AM, KaFai Wan wrote:
> OpenWRT users reported regression on ARMv6 devices after updating to latest
> HEAD, where tcpdump filter:
>
> tcpdump -i mon1 \
> "not wlan addr3 3c37121a2b3c and not wlan addr2 184ecbca2a3a \
> and not wlan addr2 14130b4d3f47 and not wlan addr2 f0f61cf440b7 \
> and not wlan addr3 a84b4dedf471 and not wlan addr3 d022be17e1d7 \
> and not wlan addr3 5c497967208b and not wlan addr2 706655784d5b"
>
> fails with warning: "Kernel filter failed: No error information"
> when using config:
>   # CONFIG_BPF_JIT_ALWAYS_ON is not set
>   CONFIG_BPF_JIT_DEFAULT_ON=y
>
> The issue arises because commits:
> 1. "bpf: Fix array bounds error with may_goto" changed default runtime to
>     __bpf_prog_ret0_warn when jit_requested = 1
> 2. "bpf: Avoid __bpf_prog_ret0_warn when jit fails" returns error when
>     jit_requested = 1 but jit fails
>
> This change restores interpreter fallback capability for BPF programs with
> stack size <= 512 bytes when jit fails.
>
> Reported-by: Felix Fietkau <nbd@nbd.name>
> Closes: https://lore.kernel.org/bpf/2e267b4b-0540-45d8-9310-e127bf95fc63@nbd.name/
> Fixes: 6ebc5030e0c5 ("bpf: Fix array bounds error with may_goto")
> Fixes: 86bc9c742426 ("bpf: Avoid __bpf_prog_ret0_warn when jit fails")
> Signed-off-by: KaFai Wan <kafai.wan@linux.dev>
> ---
>   kernel/bpf/core.c | 12 +++++++-----
>   1 file changed, 7 insertions(+), 5 deletions(-)
>
> diff --git a/kernel/bpf/core.c b/kernel/bpf/core.c
> index 5d1650af899d..2d86bd4b0b97 100644
> --- a/kernel/bpf/core.c
> +++ b/kernel/bpf/core.c
> @@ -2366,8 +2366,8 @@ static unsigned int __bpf_prog_ret0_warn(const void *ctx,
>   					 const struct bpf_insn *insn)
>   {
>   	/* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
> -	 * is not working properly, or interpreter is being used when
> -	 * prog->jit_requested is not 0, so warn about it!
> +	 * or may_goto may cause stack size > 512 is not working properly,
> +	 * so warn about it!
>   	 */
>   	WARN_ON_ONCE(1);
>   	return 0;
> @@ -2478,10 +2478,10 @@ static void bpf_prog_select_func(struct bpf_prog *fp)
>   	 * But for non-JITed programs, we don't need bpf_func, so no bounds
>   	 * check needed.
>   	 */
> -	if (!fp->jit_requested &&
> -	    !WARN_ON_ONCE(idx >= ARRAY_SIZE(interpreters))) {
> +	if (idx < ARRAY_SIZE(interpreters)) {
>   		fp->bpf_func = interpreters[idx];
>   	} else {
> +		WARN_ON_ONCE(!fp->jit_requested);
>   		fp->bpf_func = __bpf_prog_ret0_warn;
>   	}

Your logic here is to do interpreter even if fp->jit_requested is true.
This is different from the current implementation.

Also see below code:

static unsigned int __bpf_prog_ret0_warn(const void *ctx,
                                          const struct bpf_insn *insn)
{
         /* If this handler ever gets executed, then BPF_JIT_ALWAYS_ON
          * is not working properly, or interpreter is being used when
          * prog->jit_requested is not 0, so warn about it!
          */
         WARN_ON_ONCE(1);
         return 0;
}

It mentions to warn if the interpreter is being used when
prog->jit_requested is not 0.

So if prog->jit_requested is not 0, it is expected not to use interpreter.


>   #else
> @@ -2505,7 +2505,7 @@ struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
>   	/* In case of BPF to BPF calls, verifier did all the prep
>   	 * work with regards to JITing, etc.
>   	 */
> -	bool jit_needed = fp->jit_requested;
> +	bool jit_needed = false;
>   
>   	if (fp->bpf_func)
>   		goto finalize;
> @@ -2515,6 +2515,8 @@ struct bpf_prog *bpf_prog_select_runtime(struct bpf_prog *fp, int *err)
>   		jit_needed = true;
>   
>   	bpf_prog_select_func(fp);
> +	if (fp->bpf_func == __bpf_prog_ret0_warn)
> +		jit_needed = true;
>   
>   	/* eBPF JITs can rewrite the program in case constant
>   	 * blinding is active. However, in case of error during