Re: tracepoint frames

[Daniel Jacobowitz <drow@mvista.com>]

On Mon, Oct 06, 2003 at 09:59:46AM -0400, Newman, Mark (N-Superior Technical Resource Inc) wrote:
> 
> First - thanks for the response

I don't suppose you could use a mail client which quotes normally? 
It's hard to see your responses.

> -----Original Message-----
> From: Daniel Jacobowitz [mailto:drow@mvista.com]
> Sent: Monday, October 06, 2003 9:51 AM
> To: Newman, Mark (N-Superior Technical Resource Inc)
> Cc: Jim Blandy; gdb@sources.redhat.com
> Subject: Re: tracepoint frames
> 
> 
> On Mon, Oct 06, 2003 at 09:03:03AM -0400, Newman, Mark (N-Superior
> Technical Resource Inc) wrote:
> > 
> > Jim -
> > 
> > When a trace point is hit some data is collected - certainly at a
> > minimum the data specified by the collect statements.  However from
> some
> > earlier conversations and a converstaion with Ramana that additional
> > information should be collected.  Michael indicated that he collected
> a
> > "frame" in addition to the registers, data items, etc specified in the
> > collect commands.
> > 
> > Is it necessary to collect enough information to support say a
> > "backtrace" command (after a tfind)? 
> 
> Well, it would be nice but it's not generally possible.  The backtrace
> logic is pretty hairy and target-dependent; the stub has no way to find
> out what will be necessary.



> If this is necessary I was thinking that the sub could collect the whole
> stack.  However this seems to be prohibitively expensive in both size
> and speed.

Yes, it is.  On some architectures it's not even enough.

> > I have found that simple "print" commands will work and that "printf"
> > commands will not work unless one sets up the complete environment. Is
> > there a requirement or a preference on the part of the community as to
> > what needs to be available when analyzing a tracepoint?
> 
> Probably if any additional data ought to be collected that shoud be
> implemented in the GDB client, not silently by the stub.
> 
> I thought that the data was collected only in the stub when a tracepoint
> is hit.  GDB never sees the data until a "g" or an "m" protocol message
> arrives at the stub.

Right, but GDB could request more information from the stub when it
creates the tracepoint.  The stub shouldn't have to collect anything at
all that GDB didn't tell it to.


-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer