Re: [RFC PATCH] dynamic debug - adding ring buffer storage support

[Frederic Weisbecker <fweisbec@gmail.com>]

On Mon, Dec 28, 2009 at 10:24:02AM +0100, Ingo Molnar wrote:
> > that way you need to enable tracing as well... but thats ok I guess :)
> > 
> > I was investigating trace events for this, but did not find a way
> > to put variable length argument inside... and I overlooked the
> > trace_printk, I'll look on it and see how it fits, thanks
> > 
> > also having separate ring buffer makes the 'trace'/'trace_pipe' code
> > really simple (suprissingly) compared to ftrace, and I thought
> > on this place it could last for some time.. ;)
> 
> I think what we want is a unified channel of events, of which printk (and 
> dynamic-printk) is one form. I.e. we should add printk events and 
> dynamic-printk events as well, which would show up in /debug/tracing/events/ 
> in a standard ftrace event form and would be accessible to tooling that way.
> 
> For printk a single event would be enough i suspect (we dont want a separate 
> event for every printk), and for dynamic-printk we want to map the existing 
> dyn-printk topologies into /debug/tracing/events, to preserve the distinctions 
> and controls available there.
> 
> This way in the long run we'd have one unified facility.
> 
> 	Ingo


That said, I sometimes dream about one event per printk.

Having, say:

/debug/tracing/events/printk/
         |
         ---- kernel/
         |      |
         |      ------- time/
         |      |        |
         |      |        ---- clocksource.c
         |      |                 |
         |      |                 --- clocksource_unstable:218/
         |      |                 |            |
         |      |                 |            ---- format
         |      |                 |            |
         |      |                 |            ---- filter
         |      |                 |            |
         |      |                 |            ---- enable
         |      |                 --- [...]
         |      ------- [...]
         |
         ---- drivers/
         |       |
         |       ---- [...]
         |
         ---- [...]


That would give a total control over every printk, trace_printk, etc...

Too bad that would bloat the memory.
Well, that could be wrapped in a single, wildly implemented (understand:
not using TRACE_EVENT macro) trace event, something able to walk through
every calls of printk, trace_printk, early_printk, etc... and imitate
a per printk event granularity.

But still it needs to be useful...