Re: [PATCH bpf-next] bpf: remove unnecessary rcu_read_lock in kprobe_multi_link_prog_run
From: Alexei Starovoitov <hidden>
Date: 2025-08-27 16:22:03
Also in:
bpf, lkml
On Wed, Aug 27, 2025 at 5:38 AM Menglong Dong [off-list ref] wrote:
Preemption is disabled in ftrace graph, which indicate rcu_read_lock. So the rcu_read_lock is not needed in fprobe_entry(), and it is not needed in kprobe_multi_link_prog_run() neither.
kprobe_busy_begin() doing preempt_disable() is an implementation detail that might change. Having explicit rcu_read_lock() doesn't hurt. It's a nop anyway in PREEMPT_NONE. pw-bot: cr
quoted hunk ↗ jump to hunk
Signed-off-by: Menglong Dong <redacted> --- kernel/trace/bpf_trace.c | 2 -- 1 file changed, 2 deletions(-)diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c index 606007c387c5..0e79fa84a634 100644 --- a/kernel/trace/bpf_trace.c +++ b/kernel/trace/bpf_trace.c@@ -2741,12 +2741,10 @@ kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link, goto out; } - rcu_read_lock(); regs = ftrace_partial_regs(fregs, bpf_kprobe_multi_pt_regs_ptr()); old_run_ctx = bpf_set_run_ctx(&run_ctx.session_ctx.run_ctx); err = bpf_prog_run(link->link.prog, regs); bpf_reset_run_ctx(old_run_ctx); - rcu_read_unlock(); out: __this_cpu_dec(bpf_prog_active); --2.51.0