Re: [RFC][PATCHv2] ftrace interface for blktrace

[Arnaldo Carvalho de Melo <acme@redhat.com>]

Em Tue, Jan 20, 2009 at 03:58:35PM -0200, Arnaldo Carvalho de Melo escreveu:
> Em Tue, Jan 20, 2009 at 12:03:17PM -0500, Steven Rostedt escreveu:
> > > @@ -163,23 +206,33 @@ static void __blk_add_trace(struct blk_trace *bt, sector_t sector, int bytes,
> > >  
> > >  	t = relay_reserve(bt->rchan, sizeof(*t) + pdu_len);
> > 
> > Do we still use relay?
> 
> blkFtrace is an addition to the existing method for the moment, as soon
> as we're all satisfied with functionality + performance, we may then
> remove that code.

I guess that is not a clear answer, so lemme try again:

The relay code is only used if the old method of starting a blk trace
and setting up filters is used, i.e. the IOCTL method.

When we use the sysfs interface to do that we do not use the relay code
at all.

But most of the rest is shared.

Except for things like __trace_note_message and the interface provided
by the old blktrace code to insert annotations into the stream.

That will be provided by using standard ftrace methods such as
/t/trace_marker, as demonstrated by:

[root@f10-1 ~]# cat /t/trace_pipe & echo "bye, bye relay" > /t/trace_marker
[1] 1090
[root@f10-1 ~]#             bash-957   [000] 14727.302000: 0: bye, bye relay

- Arnaldo