Re: [PATCH 1/2] trace-cmd: Add new API tracecmd_set_merge_peer()

[Steven Rostedt <rostedt@goodmis.org>]

On Thu,  9 Apr 2020 16:28:34 +0300
"Tzvetomir Stoyanov (VMware)" <tz.stoyanov@gmail.com> wrote:

> +/**
> + * tracecmd_set_merge_peer - Link a tracing peer to this handle
> + * @handle: input handle for the trace.dat file
> + * @peer: input handle for the tracing peer
> + *
> + * When tracing host and one or more guest machines at the same time,
> + * guest and host are tracing peers. There is information in both trace
> + * files, related to host PID to guest vCPU mapping, timestamp synchronization
> + * and other. This information is useful when opening files at the same time and
> + * merging the events. When the host is set as a tracing peer to the guest, then
> + * the timestamps of guest's events are recalculated to match the host event's time
> + */
> +int tracecmd_set_merge_peer(struct tracecmd_input *handle,
> +			    struct tracecmd_input *peer)

I wonder it would be better to call this tracecmd_pair_peer(), like
pairing a bluetooth headset with your phone?

> +{
> +	if (!handle)
> +		return -1;
> +

This should probably fail if the host.peer_data is already set. Which
means we should also have a way to unmerge (unpair) the two. Which is
why I like the term "pair" better, as it is like paring a bluetooth
device and then unpairing it.

-- Steve


> +	handle->host.peer_data = peer;
> +	tracecmd_ref(peer);
> +	tsync_check_enable(handle);
> +
> +	return 0;
> +}
> +
>  /**
>   * tracecmd_ref - add a reference to the handle
>   * @handle: input handle for the trace.dat file
> @@ -3785,7 +3839,7 @@ int tracecmd_get_guest_cpumap(struct tracecmd_input *handle,
>   */
>  unsigned long long tracecmd_get_tsync_peer(struct tracecmd_input *handle)
>  {
> -	return handle->host.trace_id;
> +	return handle->host.peer_trace_id;
>  }
>