Re: Unified tracing buffer

["K.Prasad" <prasad@linux.vnet.ibm.com>]

On Tue, Sep 23, 2008 at 07:00:38AM -0700, Martin Bligh wrote:
> > - get rid of anything having to do with padding, nobody needs it and its
> > only affect has been to horribly distort and complicate a lot of the
> > code
> > - get rid of sub-buffers, they just cause confusion
> > - get rid of mmap, nobody uses it
> > - no sub-buffers and no mmap support means we can get rid of most of the
> > callbacks, and a lot of API confusion along with them
> > - add relay flags - they probably should have been used from the
> > beginning and options made explicit instead of being shoehorned into the
> > callback functions.
> 
> Actually, I think if you did all that, it'd be pretty close to what we
> want anyway ...

In the perspective of having a layered infrastructure, can we consider
the interfaces later added over relay (to be used as a wrapper), namely
relay_printk() and relay_dump()?

Also add the following features to it and we get close to the
functionality that is sought:

- Add callbacks to append fine-granular timestamp information depending
upon user's requirement
- Ability to provision more custom-defined control files that can suit
independent tracer's requirements

These interfaces already come along with the following features (some
repetition here from my previous email for the sake of completeness):

- Very minimal work required to log data using the interfaces. Usage is
made simple to resemble the printk(). Like

	struct relay_printk_data *tpk;
	tpk->parent_dir = "PARENT";
	tpk->dir = "DIR";
	relay_printk(tpk, <String to be output>);
	relay_dump(tpk, <Some binary data to output>); 

Output at:
	<debugfs_mount>/PARENT/DIR/<TRACE FILES>

- Assumes default values for most tunables, such as per-CPU buffer size,
relay-flags (such as global vs local per-cpu buffers, flight recorder vs
overwrite mode), thus reducing the work required for setting up these
interfaces. They can be over-written in case of advanced needs.

- Well defined control operations to start, stop tracing operations, status
files to indicate buffer overflow, etc.

- Given the recent patches that Tom Zanussi has sent to bring in the
erstwhile 'trace' functionality into relay itself, there can be a lot of
code-reduction in relay.c (in -mm) thereby leading to a light-weight
implementation.

Thanks,
K.Prasad