Re: [PATCH bpf-next 1/4] bpf: Add bpf_call_rcu() kfunc

[sashiko-bot@kernel.org]

Thank you for your contribution! Sashiko AI review found 1 potential issue(s) to consider:
- [High] Calling `bpf_call_rcu` from NMI context causes a raw spinlock deadlock because `call_rcu()` is not NMI-safe.
--

commit 3385037afe11a72dffbb6db99b4174626596c231
Author: Puranjay Mohan <puranjay@kernel.org>

bpf: Add bpf_call_rcu() kfunc

This commit adds a new bpf_call_rcu kfunc that allows BPF programs to defer
operations using RCU grace periods.

> diff --git a/kernel/bpf/helpers.c b/kernel/bpf/helpers.c
> --- a/kernel/bpf/helpers.c
> +++ b/kernel/bpf/helpers.c
[ ... ]
> +__bpf_kfunc int bpf_call_rcu(struct bpf_rcu_head *rh, void *map__const_map,
> +			     bpf_rcu_callback_t callback, struct bpf_prog_aux *aux)
> +{
[ ... ]
> +	rhk->callback_fn = (bpf_callback_t)(void *)callback;
> +	rhk->map = map;
> +	rhk->prog = prog;
> +	call_rcu(&rhk->rcu, bpf_rcu_run_callback);

[Severity: High]
Does this code risk a raw spinlock deadlock if called from an NMI context?

Because bpf_call_rcu is exposed to tracing programs, it can be invoked from
NMI contexts, such as when a tracing program is attached to a perf event.
However, call_rcu acquires rcu_node raw spinlocks and is not NMI-safe. If
the NMI interrupts a path that already holds the same rcu_node lock, the
system will attempt to re-acquire the lock, resulting in a deadlock.

Should the verifier restrict this kfunc from being called in NMI contexts,
similar to how it uses btf_record_has_nmi_unsafe_fields for bpf_obj_drop?

-- 
Sashiko AI review · https://sashiko.dev/#/patchset/20260907134552.1772405-1-puranjay@kernel.org?part=1