Re: [PATCH v2] KASAN: prohibit KASAN+STRUCTLEAK combination

[Dennis Zhou <dennisszhou@gmail.com>]

Hi,

On Fri, Apr 20, 2018 at 07:56:56AM +0200, Dmitry Vyukov wrote:
> As a sanity check, I would count number of zeroing inserted by the
> plugin it both cases and ensure that now it does not insert order of
> magnitude more/less. It's easy with function calls (count them in
> objdump output), not sure what's the easiest way to do it for inline
> instrumentation. We could insert printf into the pass itself, but it
> if runs before inlining and other optimization, it's not the final
> number.

I modified the structleak_plugin to count the number of initializations
and output if the function was an inline function or not. The aggregated
values are below.

declared inline       no       yes
----------------------------------
early_optimizations:  12168   7114
*all_optimizations:   12554     13

These numbers seem appropriate. The structleak initializes in declared
inline functions are redundant.

> Also note that asan pass is at different locations in the pipeline
> depending on optimization level:
> https://gcc.gnu.org/viewcvs/gcc/trunk/gcc/passes.def?view=markup

The *all_optimizations pass happens before any of the asan pass
locations so I think this shouldn't change those semantics.

Thanks,
Dennis