|
From: Jeremy F. <je...@go...> - 2004-03-13 20:30:01
|
On Sat, 2004-03-13 at 10:24, Tom Hughes wrote:
> I've been building calltree for 2.1.0 with only a minor patch to
> accept the newer interface version, and I did a patch this morning
> to make it work with 2.1.1 which is attached.
>
> The only problem I've notice with 2.1.1 with this patch is the need
> to use pointercheck=no as Josef has mentioned on the mailing list a
> couple of times.
Yeah, I'm not quite sure what the right fix is for that. It seems to me
the two options are:
1. add new CLOAD/CSTORE (checked and/or client load and store),
which is only generated by vg_to_ucode; the tools would use
LOAD/STORE for their own instrumentation memory accesses
2. add a flag to the existing LOAD and STORE
I'm tending towards option 1, but since we're planning on changing
everything for multi-arch stuff, I'm not inclined to spend to much time
on it now.
In the meantime, the quick workaround for calltree is for it to clear
VG_(clo_pointercheck) in its post-CLO parsing init function.
J
|