Re: [RFC PATCH 1/1] tracepoints: tree-wide: Replace %p with %px

[Masami Hiramatsu]

On Wed, 14 Oct 2020 11:35:17 -0400
Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:

> On Thu, 15 Oct 2020 00:11:04 +0900
> Masami Hiramatsu <mhiramat@xxxxxxxxxx> wrote:
> 
> > On Wed, 14 Oct 2020 09:38:13 -0400
> > Steven Rostedt <rostedt@xxxxxxxxxxx> wrote:
> > 
> > 
> > Hmm, would you mean we always run such conversion on printing the trace
> > buffer for each entry? It could be much overhead because we need allocate
> > memory (%p->%px increase 1 byte) and format conversion (with copying it).
> 
> I was thinking of having a single buffer allocated to the iterator (could
> increase in size when needed). Really, the output isn't that time critical,
> and running a conversion on all events each time shouldn't be that bad. And
> if it is, we could have a cache in the iter for a set of events to handle.
> Note, the conversion only needs to happen for events that have a %p in the
> fmt, and only those need to be cached.
> 
> Then those buffers could be freed when the iter is freed.

Ah, I got it. so you meant adding a buffer field to struct trace_iterator,
correct?

> 
> > Maybe we can avoid repeating it using a kind of cache, but it also consumes
> > memory. And as I pointed, the security reason is meaningless because there
> > are raw addresses in raw data which user can read...
> > Could you tell me what is your point? Making the code change as small as
> > possible?
> > 
> 
> It's about not having to worry about this in the future. Otherwise, we'll
> be playing whack-a-mole on making all %px in new trace events.

Hmm, OK. I hope no one adds an event which shows both ptr and hashed ptr
for making a database in the future ;)
Let me try to re-implement it.

Thank you,

-- 
Masami Hiramatsu <mhiramat@xxxxxxxxxx>