Re: DTrace of radeonkms on 9.1

On Mon, 1 Apr 2013 14:36:52 +0200 Alexander Leidinger <Alexander_at_leidinger.net> wrote:
>
> On Wed, 27 Mar 2013 18:07:14 -0400
> "J.R. Oldroyd" <fbsd_at_opal.com> wrote:
> 
> > Is there any known magic involved in getting DTrace to do its thing on
> > 9.1-release?
> > 
> > I am trying to use it to debug a memory leak problem with the
> > radeonkms driver under 9.x.
> 
> Can you check if you have the same dtrace-problem with -current? I
> would expect that 9.1 already has some dtrace-fixes regarding new
> probes in run-time loaded modules, this is just to verify this
> assumption.
> 
> Assuming there is some kind of dead-lock in this module-load interaction
> with dtrace, you could modify the radeonkms module to do it's
> initialization magic once a sysctl is set to 1, instead of doing this
> magic at module-load time. This way you could load the module, start
> the dtrace script and then issue the magic sysctl.
> 
> Bye,
> Alexander.
> 

Thanks for the suggestion.

The problem I was hoping to use DTrace to debug turned out to be
an incompletely back-ported vm_alloc_page_contig() function.  This
was discovered using careful code analysis.  Having improved the
back-port of this function, the radeonkms driver is now running
fine on 9.1-release.  The need for DTtrace has therefore gone away
for the time being.

That said, the back-port of the VM function is not 100%, yet.
I've had to put this on hold due to other commitments, but when
I get back to this, I probably need to talk to our VM experts to
get a list of commits that will be needed to properly merge that
function into 9-stable.

	-jr