Messages in this thread | | | Date | Fri, 6 Dec 2013 15:21:41 +0000 | From | Dave Martin <> | Subject | Re: [PATCH V4] ARM : unwinder : Prevent data abort due to stack overflow |
| |
On Thu, Dec 05, 2013 at 03:15:22PM +0000, Anurag Aggarwal wrote: > >>if (unlikely(check) && > >> *vsp >= (unsigned long *)ctrl->sp_high)) > > Good idea, It can help in optimizing the code and will leave code more readable > and it will be easy to maintain also.
That was my goal -- it's not always easy to achieve both, but we should try where possible, especially for non-performance-critical code.
> >>Does the code check anywhere that SP is word-aligned? > >> > >>That should probably be checked if it's not done already. > > I think this should be handled in separate patch > I would also like to hear more your ideas for the file
There are some limitations on how backtracing works with ARM_UNWIND. The backtracer interface assumes that the compiler's register use with ARM_UNWIND is "almost the same" as with !ARM_UNWIND. But this is not always true, and we've seen some problems.
I think the best solution will involve some refactoring of the interface between the backtracer and the unwinder backends, but I don't have it all figured out yet...
Cheers ---Dave
| |