[cfe-commits] [llvm-commits] [PATCH] [asan] Use zero-based shadow by default on Android

[Evgeniy Stepanov]

r157923

On Mon, Jun 4, 2012 at 1:44 PM, Evgeniy Stepanov
<eugeni.stepanov at gmail.com> wrote:
> On Fri, May 25, 2012 at 3:08 PM, Chandler Carruth <chandlerc at google.com> wrote:
>> On Fri, May 25, 2012 at 3:16 AM, Evgeniy Stepanov
>> <eugeni.stepanov at gmail.com> wrote:
>>>
>>> On Wed, May 23, 2012 at 4:07 PM, Evgeniy Stepanov
>>> <eugeni.stepanov at gmail.com> wrote:
>>> > Good point. Both patches landed, here is another one that adds a
>>> > driver check for -pie.
>>>
>>> ping w/ a patch
>>
>>
>> LGTM
>>
>> The diagnostic says that '-pie' is required with ASan on Android, but
>> nothing about during compiling vs. during linking? Do we blanket require
>> this flag, or does it only apply to some steps? Might be nice to make the
>> diagnostic explicit here.
>
> -pie only seems to affect linking. We don't really care about
> compilation step, as long as linking produces a position-independent
> executable.
>
>
>>
>>>
>>>
>>>
>>> > On Wed, May 23, 2012 at 3:04 PM, Chandler Carruth <chandlerc at google.com>
>>> > wrote:
>>> >> On Wed, May 23, 2012 at 3:58 AM, Evgeniy Stepanov
>>> >> <eugeni.stepanov at gmail.com> wrote:
>>> >>>
>>> >>> Note that this change will require building binaries with -fPIE -pie.
>>> >>> I believe there is really no way around this. We can not change shadow
>>> >>> location on a per-binary basic because of the shared runtime, and
>>> >>> there is no universal location that fit both PIE and non-PIE.
>>> >>>
>>> >>> I'm going to land this soon, then add some runtime checks to ensure
>>> >>> that the shadow does not silently override any preexisting mappings.
>>> >>
>>> >>
>>> >> I haven't looked at the patches, but it would be good to also teach the
>>> >> Clang driver to produce an error if the appropriate options are not
>>> >> attached
>>> >> to the compile and link commands but '-faddress-sanitizer' is.
>>> >>
>>> >>>
>>> >>>
>>> >>> On Tue, May 22, 2012 at 6:43 PM, Kostya Serebryany <kcc at google.com>
>>> >>> wrote:
>>> >>> > Both look good, thanks!
>>> >>> > --kcc
>>> >>> >
>>> >>> > On Tue, May 22, 2012 at 6:38 PM, Evgeniy Stepanov
>>> >>> > <eugeni.stepanov at gmail.com> wrote:
>>> >>> >>
>>> >>> >> Hi,
>>> >>> >>
>>> >>> >> these patches (one to llvm, another to compiler-rt) change ASan
>>> >>> >> default shadow offset on Android to zero. This seems to marginally
>>> >>> >> improve performance of Thumb code, significantly (~5%) saves code
>>> >>> >> size, and avoids address space collisions with the main executable
>>> >>> >> in
>>> >>> >> a non-ASLR case.
>>> >>> >>
>>> >>> >> Please review.
>>> >>> >>
>>> >>> >> _______________________________________________
>>> >>> >> llvm-commits mailing list
>>> >>> >> llvm-commits at cs.uiuc.edu
>>> >>> >> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>> >>> >>
>>> >>> >
>>> >>> _______________________________________________
>>> >>> llvm-commits mailing list
>>> >>> llvm-commits at cs.uiuc.edu
>>> >>> http://lists.cs.uiuc.edu/mailman/listinfo/llvm-commits
>>> >>
>>> >>
>>
>>