Re: [PATCH 0/3] Stack checking on Arm

[Stefano Stabellini <sstabellini@xxxxxxxxxx>]

On Mon, 29 Jul 2024, Julien Grall wrote:
> Hi,
> 
> On 29/07/2024 15:24, Stewart Hildebrand wrote:
> > This series introduces stack check instrumentation on Arm. This is
> > helpful for safety certification efforts. I'm sending this in an RFC
> > state because I wanted to gather opinions on the approach of using
> > -finstrument-functions.
> 
> This looks ok for an initial approach. I wonder if longer term we want to
> implement stack guards on Arm. We would need to allocate an extra "virtual"
> page per stack that would be unmapped.
> 
> The advantage is this could be used also in production and doesn't rely on any
> support from the processor.
> 
> Any thoughts?

I think we need both. We should implement stack guards on Arm. In
addition, it is also beneficial to have -finstrument-functions for
profiling, debugging, and also so that we can retrieve detailed call
graphs from execution runs. As an example, -finstrument-functions can
help with offline analysis to prove that we don't have unbounded
recursion, on both arm and x86 too. On the other hand, stack guards help
with protecting the stack in production.