On Thu, May 05, 2022 at 01:45:37PM +0100, Mark Rutland wrote:
> Hi Kees,
> 
> I hadn't realised the LKDTM STACKLEAK test could be built with
> CONFIG_STACKLEAK=n, and my rework of that depends upon helpers that only exist
> when CONFIG_STACKLEAK=y.
> 
> IMO the test is nonsensical for CONFIG_STACKLEAK=n, and I reckon we should
> either:
> 
> a) Not build the stackleak test at all when CONFIG_STACKLEAK=n
> 
> b) Have a small stub that just logs that CONFIG_STACKLEAK=n and the test is
>    being skipped.
> 
> Do you have any preference between the two?

Since it's looking for a specific poison, it doesn't make sense to look
for this property as magically appearing (where as this kind of thing
sometimes exists for other tests: did the hypervisor block it instead of
the kernel, etc)

So, yeah, I'd wrap it in an ifdef with an else: XFAIL.

-- 
Kees Cook