Re: [PATCH v13 11/11] arm64: Select HAVE_RELIABLE_STACKTRACE

["Madhavan T. Venkataraman" <madvenka@linux.microsoft.com>]

On 1/26/22 04:20, nobuta.keiya@fujitsu.com wrote:
>> I have not seen any activity on that in a long time. IIRC, Julien quit RedHat.
>> I don't know if anyone else has taken over this work in RedHat.
>>
>> Sorry, I don't have any more information.
>>
>> Madhavan
> 
> Thanks for your information.
> 
> By the way, I'm considering test code for arch_stack_walk_reliable().
> Specifically, I apply Suraj's patch to enable livepatch, and added a function
> that sleeps between SYM_CODE_START and SYM_CODE_END, then livepatch
> checks if the task has an unreliable stack.
> For now my internal test code working correctly, but my Kconfig excludes
> STACK_VALIDATION dependency.
> 
> It seems that objtool will not be enabled yet, so I would like to test it easier.
> If you are already testing with this patch, could you tell me how to do it?
> 
> 

For now, I have an instrumented kernel that directly invokes arch_stack_walk_reliable()
from various places in the kernel (interrupt handlers, exception handlers, ftrace entry,
kprobe handler, etc). I also have a test driver to induce conditions like null pointer
dereference. I use this to test different cases where arch_stack_walk_reliable() should
return an error.

As for livepatch testing, I have enhanced objtool and the kernel so the frame pointer can
be validated dynamically rather than statically. I have tested various different livepatch
selftests successfully. I have also written my own livepatch tests to add to the selftests.
I am currently working on preparing an RFC patch series for review. Basically, this series
implements STACK_VALIDATION in a different way.

I plan to publish my work soon (hopefully Feb 2022). I was going to do in December. However,
my workload in Microsoft did not permit me to do that. I am also planning to set up a github
repo so people can try out my changes, if they are interested.

So, stay tuned.

Madhavan