On Thu, Sep 20, 2007 at 09:22:36AM -0700, michael schuster wrote:
> Mikael Kjerrman wrote:
> > I'm answering the first one myself since I actually found it by
> > searching. So 'dtrace -v' gives me a hint about the arguments. But as
> > for my second question, I have still not been able to decode filesystem
> > names, directories etc. when vxfs is involved. Has anyone out there been
> > successful with dtracing vxfs?
>
> IIRC someone once mentioned that that's not possible with vxfs (I assume
> due to the way they compile their stuff).
Fortunately, the only to _compile_ your stuff such to completely evade
DTrace would be to either invent your own ABI or write your own kernel
linker -- and you would probably need to do both. The only thing that
I can think of that you might be making reference to is that on SPARC,
our FBT instrumentation methodology (namely, branch insertion) requires
that modules are no larger than the size of the branch displacement minus
the size of the VA that we reserve for our trampoline. Upshot: we
can't instrument modules on SPARC that have more than 3MB of text. This
is very rare, and (frankly) reflects dubious software engineering, but
to see if you're falling into this, run "size -f" on the kernel binary
(e.g. "size -f /kernel/fs/vxfs"). And if you are running into this,
you might want to ask Symantec why it's taking them more than 3MB to
do less than what ZFS does in ~400K... ;)
- Bryan
--------------------------------------------------------------------------
Bryan Cantrill, Solaris Kernel Development. http://blogs.sun.com/bmc
_______________________________________________
dtrace-discuss mailing list
[email protected]
