Re: [PATCH bpf-next v8 2/2] bpf: Reject programs with inlined helpers if JIT is not available

[Eduard Zingerman <eddyz87@gmail.com>]

On Tue, 2026-07-07 at 18:39 +0800, Tiezhu Yang wrote:

[...]

> Hi Eduard,
> 
> Sorry for the misunderstanding.
> Your wrapper handles the load-time JIT-disabled case nicely.
> 
> The problem is, it cannot protect against the case where JIT
> is enabled during verification but dynamically fails later.
> 
> In that scenario, the verifier would have already skipped the
> rewrite based on the wrapper, and the fallback to interpreter
> would still trigger the kernel panic.

I'm not sure I understand what you mean. The decision on whether to
use jit for a specific program should be taken at the beginning of the
load system call processing. bpf_prog_alloc_no_stats() sets the
fp->jit_requested before the bpf_check() call in core.c:bpf_prog_load()
and that should be a final say.

At no point the program should be processed with bpf_check() assuming
that jit is enabled and runtime is selected assuming that jit is
disabled. If there are such instances now, these should be addressed.

> By using the 'jit_required' flag, we can cleanly catch both
> cases, and reject early or return -ENOTSUPP at runtime.
> 
> Please see the test log for the detailed fallback behaviors:
> https://lore.kernel.org/loongarch/aa26b3fc-9717-4273-dd76-a0dd6b15ff61@loongson.cn/
> 
> Since you mentioned it "might as well land as-is", I will
> proceed with v9 using this flag architecture.
> 
> Thanks,
> Tiezhu