Messages in this thread | | | Date | Tue, 24 Mar 2009 23:14:39 +0100 | From | Ingo Molnar <> | Subject | Re: Anyone working on ftrace function graph support on ARM? |
| |
* Frederic Weisbecker <fweisbec@gmail.com> wrote:
> On Tue, Mar 24, 2009 at 10:48:46PM +0100, Ingo Molnar wrote: > > > > * Frederic Weisbecker <fweisbec@gmail.com> wrote: > > > > > Well it's a very naive listing, there are sometimes some problems. > > > For example on x86-64, I had to save even some non-scratch > > > registers before calling the return hook, I still don't know why. > > > > btw., which are those registers? > > > > Ingo > > > I would expect to only save rax,rdi,rsi,rdx,rcx,r8,r9 which are > used for parameters.
> And I had some crashes until I append r10 and r11 which actually > are scratch if I'm not wrong, but since they are scratch and are > not used for arguments, I thought they didn't need to be saved. > > Well, I think there were some code flow cases I was missing.
Correct, r10 and r11 are clobbered registers too - and you need to save them too in mcount methods.
The reason is that mcount has a special calling convention - it's not just about not destroying arguments - GCC can keep data in r10 or r11 scratch registers across function calls as well - for example for relatively static functions that are in its local optimization scope.
If GCC can prove that the local scope function itself does not clobber r10/r11, it does not have to clobber them across the function call. But the mcount() callback still gets inserted.
So the rule is: mcount must not destroy _any_ register state. (beyond flags)
ngo
| |