On Mon, Aug 30, 2021 at 2:42 PM Song Liu <songliubraving@xxxxxx> wrote:
>
> Introduce bpf_get_branch_snapshot(), which allows tracing pogram to get
> branch trace from hardware (e.g. Intel LBR). To use the feature, the
> user need to create perf_event with proper branch_record filtering
> on each cpu, and then calls bpf_get_branch_snapshot in the bpf function.
> On Intel CPUs, VLBR event (raw event 0x1b00) can be use for this.
>
> Signed-off-by: Song Liu <songliubraving@xxxxxx>
> ---
> include/linux/bpf.h | 2 ++
> include/linux/filter.h | 3 ++-
> include/uapi/linux/bpf.h | 16 +++++++++++++
> kernel/bpf/trampoline.c | 13 ++++++++++
> kernel/bpf/verifier.c | 12 ++++++++++
> kernel/trace/bpf_trace.c | 43 ++++++++++++++++++++++++++++++++++
> tools/include/uapi/linux/bpf.h | 16 +++++++++++++
> 7 files changed, 104 insertions(+), 1 deletion(-)
>
[...]
> diff --git a/kernel/bpf/verifier.c b/kernel/bpf/verifier.c
> index 206c221453cfa..72e8b49da0bf9 100644
> --- a/kernel/bpf/verifier.c
> +++ b/kernel/bpf/verifier.c
> @@ -6446,6 +6446,18 @@ static int check_helper_call(struct bpf_verifier_env *env, struct bpf_insn *insn
> env->prog->call_get_func_ip = true;
> }
>
> + if (func_id == BPF_FUNC_get_branch_snapshot) {
> + if (env->prog->aux->sleepable) {
> + verbose(env, "sleepable progs cannot call get_branch_snapshot\n");
> + return -ENOTSUPP;
> + }
> + if (!IS_ENABLED(CONFIG_PERF_EVENTS)) {
> + verbose(env, "func %s#%d not supported without CONFIG_PERF_EVENTS\n",
> + func_id_name(func_id), func_id);
> + return -ENOTSUPP;
> + }
> + env->prog->call_get_branch = true;
> + }
> if (changes_data)
> clear_all_pkt_pointers(env);
> return 0;
> diff --git a/kernel/trace/bpf_trace.c b/kernel/trace/bpf_trace.c
> index 8e2eb950aa829..a01f26b7877e6 100644
> --- a/kernel/trace/bpf_trace.c
> +++ b/kernel/trace/bpf_trace.c
> @@ -1017,6 +1017,33 @@ static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
> .arg1_type = ARG_PTR_TO_CTX,
> };
>
> +BPF_CALL_2(bpf_get_branch_snapshot, void *, buf, u32, size)
I bet we'll need u64 flags over time, let's add it right now. It's
similar to bpf_read_branch_records().
> +{
> +#ifdef CONFIG_PERF_EVENTS
> + u32 max_size;
> +
> + if (this_cpu_ptr(&bpf_perf_branch_snapshot)->nr == 0)
> + return -EOPNOTSUPP;
> +
> + max_size = this_cpu_ptr(&bpf_perf_branch_snapshot)->nr *
> + sizeof(struct perf_branch_entry);
> + memcpy(buf, this_cpu_ptr(&bpf_perf_branch_snapshot)->entries,
> + min_t(u32, size, max_size));
> +
Check bpf_read_branch_records() implementation and it's argument
validation logic. Let's keep them consistent (e.g., it enforces that
size is a multiple of sizeof(struct perf_branch_entry)). Another
difference is that bpf_read_branch_records() returns number of bytes
filled, not number of records. That's consistent with accepting size
as number of bytes. Let's stick to this convention then, so bytes
everywhere.
> + return this_cpu_ptr(&bpf_perf_branch_snapshot)->nr;
> +#else
> + return -EOPNOTSUPP;
> +#endif
> +}
> +
> +static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
> + .func = bpf_get_branch_snapshot,
> + .gpl_only = true,
> + .ret_type = RET_INTEGER,
> + .arg1_type = ARG_PTR_TO_UNINIT_MEM,
> + .arg2_type = ARG_CONST_SIZE_OR_ZERO,
> +};
> +
[...]
