On 11/23/2017 03:01 PM, Michal Hocko wrote: > I am not sure adding a probe on a production system will fly in many > cases. A static tracepoint would be much easier in that case. But I > agree there are other means to accomplish the same thing. My main point > was to have an easy out-of-the-box way to check latencies. But that is > not something I would really insist on. > In android tracefs (or part of it) is the way for the system to control to what developers can access to the linux system on commercial devices. So it is very much used on production systems, it is even a requirement from google to be certified as android. Things like dmesg is not. However, this probe is at the moment not in that scope.
My point is that you need to condense the information as much as possible but still be useful before making the effort to copy it to userspace. And for this the trace-event are very useful for small systems since the cost is very low for events where no one is listening.
