cgroup_hierarchical_stats triggers a lockdep warning in __css_rstat_updated() on PREEMPT_RT. The counter runs at the cgroup_attach_task tracepoint under spin_lock_irqsave() on trace_cgroup_path_lock. On PREEMPT_RT this leaves preemption and interrupts enabled. On non-RT it masks interrupts, satisfying the assertion. Disable preemption around css_rstat_updated() so the updater can queue the update without being preempted. Fixes: 36df6e3dbd7e ("cgroup: make css_rstat_updated nmi safe") Reported-by: Jiayuan Chen Closes: https://lore.kernel.org/r/20251119111402.153727-1-jiayuan.chen@linux.dev/ Link: https://lore.kernel.org/r/fmey2ektwdk7l5aktjz3xb24x3pk3lxy3gfld7reazal6xa6u2@tmryedehqhql/ Signed-off-by: Karl Mehltretter --- Assisted by an LLM. Ran test_progs -t cgroup_hierarchical_stats on bpf-next 9867c61abbd4 in x86-64 KVM guests with lockdep. The test passed on all eight boots. Boots with an rstat warning Before After PREEMPT_RT 2/2 0/2 PREEMPT (!RT) 0/2 0/2 tools/testing/selftests/bpf/progs/cgroup_hierarchical_stats.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/tools/testing/selftests/bpf/progs/cgroup_hierarchical_stats.c b/tools/testing/selftests/bpf/progs/cgroup_hierarchical_stats.c index 8fc38592a..d4babda81 100644 --- a/tools/testing/selftests/bpf/progs/cgroup_hierarchical_stats.c +++ b/tools/testing/selftests/bpf/progs/cgroup_hierarchical_stats.c @@ -6,6 +6,7 @@ #include #include #include +#include "bpf_experimental.h" char _license[] SEC("license") = "GPL"; @@ -76,7 +77,9 @@ int BPF_PROG(counter, struct cgroup *dst_cgrp, const char *path, else if (create_percpu_attach_counter(cg_id, 1)) return 0; + bpf_preempt_disable(); css_rstat_updated(&dst_cgrp->self, bpf_get_smp_processor_id()); + bpf_preempt_enable(); return 0; } base-commit: 9867c61abbd44f7053a871566b3d982f62c6d8dc -- 2.43.0