Re: [PATCH v1 1/2] perf lock contention: Support pre-5.14 kernels

From: Namhyung Kim
Date: Mon Apr 10 2023 - 11:34:09 EST


Hi Ian,

On Fri, Apr 7, 2023 at 10:52 PM Ian Rogers <irogers@xxxxxxxxxx> wrote:
>
> struct rq's variable __lock was renamed from lock in 5.14.
>
> Signed-off-by: Ian Rogers <irogers@xxxxxxxxxx>

Thanks for fixing this.

Acked-by: Namhyung Kim <namhyung@xxxxxxxxxx>

Thanks,
Namhyung


> ---
> tools/perf/util/bpf_skel/lock_contention.bpf.c | 15 ++++++++++++++-
> 1 file changed, 14 insertions(+), 1 deletion(-)
>
> diff --git a/tools/perf/util/bpf_skel/lock_contention.bpf.c b/tools/perf/util/bpf_skel/lock_contention.bpf.c
> index 23f6e63544ed..8911e2a077d8 100644
> --- a/tools/perf/util/bpf_skel/lock_contention.bpf.c
> +++ b/tools/perf/util/bpf_skel/lock_contention.bpf.c
> @@ -418,6 +418,14 @@ int contention_end(u64 *ctx)
>
> extern struct rq runqueues __ksym;
>
> +struct rq__old {
> + raw_spinlock_t lock;
> +} __attribute__((preserve_access_index));
> +
> +struct rq__new {
> + raw_spinlock_t __lock;
> +} __attribute__((preserve_access_index));
> +
> SEC("raw_tp/bpf_test_finish")
> int BPF_PROG(collect_lock_syms)
> {
> @@ -426,11 +434,16 @@ int BPF_PROG(collect_lock_syms)
>
> for (int i = 0; i < MAX_CPUS; i++) {
> struct rq *rq = bpf_per_cpu_ptr(&runqueues, i);
> + struct rq__new *rq_new = (void *)rq;
> + struct rq__old *rq_old = (void *)rq;
>
> if (rq == NULL)
> break;
>
> - lock_addr = (__u64)&rq->__lock;
> + if (bpf_core_field_exists(rq_new->__lock))
> + lock_addr = (__u64)&rq_new->__lock;
> + else
> + lock_addr = (__u64)&rq_old->lock;
> lock_flag = LOCK_CLASS_RQLOCK;
> bpf_map_update_elem(&lock_syms, &lock_addr, &lock_flag, BPF_ANY);
> }
> --
> 2.40.0.577.gac1e443424-goog
>