Re: [PATCH] user_events: Enable user processes to create and write to trace events

[Masami Hiramatsu]

On Wed, 13 Oct 2021 11:40:34 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Thu, 14 Oct 2021 00:21:32 +0900
> Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
> 
> > > This approach requires an FD returned and either an int for the status
> > > page or the returend FD could expose the ID via another IOCTL being
> > > issued.  
> > 
> > OK, I would like to suggest you to add events/user-events/*/marker file
> > (which returns that shared file struct backed FD) so that some simple
> > user scripts can also send the events (these may not use ioctl, just
> > write the events.) But this can be done afterwards anyway.
> > 
> 
> I'd prefer we avoid this. It will break some of the semantics of the events
> directory. One, only "user-events" will have this "marker" file.

Yes, it is only for the user-events.

> Although
> it will be very similar to the "inject" file, but that's only for debugging
> anyway.

Oh, do we already the "inject" file?

> All the files in the events directory is starting to add a bunch of
> overhead, as they are typically copied into the instances. Although, when
> we get the eventfs directory created, maybe that will not be as big of a
> deal. But that still doesn't fix the semantics issue.

Indeed. OK, making "marker" file for each instances may confuse user because
the written event itself must be delivered to any instance but the file
seems to belong to one of them. Please ignore it.

Thank you,

> 
> -- Steve
> 


-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>