Refactor poisoning done inside ASan RTL

[kcc, 2011-11-29 17:51:10]

LGTM++
(assuming there is not change in functionality, just refactoring)

http://codereview.appspot.com/5448043/diff/1/asan/asan_mapping.h
File asan/asan_mapping.h (right):

http://codereview.appspot.com/5448043/diff/1/asan/asan_mapping.h#newcode94
asan/asan_mapping.h:94: static inline bool AddrIsAligned(uintptr_t a) {
AddrIsAlignedByGranularity

[samsonov, 2011-11-29 19:11:29]

On Tue, Nov 29, 2011 at 9:51 PM,  <konstantin.s.serebryany@gmail.com> wrote:
> LGTM++
> (assuming there is not change in functionality, just refactoring)

More checks show something interesting:

$ make b32
$ cd bin_Linux/
$ ASAN_OPTIONS="report_globals=2" ./asan_test32

<....>
Added Global: beg=0x0819c3ec size=16 name=_ZTSN7testing4TestE
==21395== CHECK failed: AddrIsAligned(g->beg) <...>

Interesting, I believed all globals are at least 32-bytes aligned.
Will take a closer look tomorrow.

>
> http://codereview.appspot.com/5448043/diff/1/asan/asan_mapping.h
> File asan/asan_mapping.h (right):
>
> http://codereview.appspot.com/5448043/diff/1/asan/asan_mapping.h#newcode94
> asan/asan_mapping.h:94: static inline bool AddrIsAligned(uintptr_t a) {
> AddrIsAlignedByGranularity
>
> http://codereview.appspot.com/5448043/
>



-- 
Alexey Samsonov
Software Engineer, Moscow
samsonov@google.com

[kcc, 2011-11-29 19:21:38]

Do you have a fresh clang binary?
This sounds like a bug I fixed recently (should not instrument odr linkage
globals).

On Tue, Nov 29, 2011 at 11:11 AM, Alexey Samsonov <samsonov@google.com>wrote:

> On Tue, Nov 29, 2011 at 9:51 PM,  <konstantin.s.serebryany@gmail.com>
> wrote:
> > LGTM++
> > (assuming there is not change in functionality, just refactoring)
>
> More checks show something interesting:
>
> $ make b32
> $ cd bin_Linux/
> $ ASAN_OPTIONS="report_globals=2" ./asan_test32
>
> <....>
> Added Global: beg=0x0819c3ec size=16 name=_ZTSN7testing4TestE
> ==21395== CHECK failed: AddrIsAligned(g->beg) <...>
>
> Interesting, I believed all globals are at least 32-bytes aligned.
> Will take a closer look tomorrow.
>
> >
> > http://codereview.appspot.com/5448043/diff/1/asan/asan_mapping.h
> > File asan/asan_mapping.h (right):
> >
> >
> http://codereview.appspot.com/5448043/diff/1/asan/asan_mapping.h#newcode94
> > asan/asan_mapping.h:94: static inline bool AddrIsAligned(uintptr_t a) {
> > AddrIsAlignedByGranularity
> >
> > http://codereview.appspot.com/5448043/
> >
>
>
>
> --
> Alexey Samsonov
> Software Engineer, Moscow
> samsonov@google.com
>

[samsonov, 2011-11-30 12:35:41]

> Do you have a fresh clang binary?

Indeed, clang revision was not bumped.

> This sounds like a bug I fixed recently (should not instrument odr linkage
> globals).

I checked out compiler-rt/lib/asan and tested a patch (setting CLANG_BUILD to
the patched Clang from ASan repository). Sent it to llvm-comits as well.


>
>
> On Tue, Nov 29, 2011 at 11:11 AM, Alexey Samsonov <samsonov@google.com>
> wrote:
>>
>> On Tue, Nov 29, 2011 at 9:51 PM,  <konstantin.s.serebryany@gmail.com>
>> wrote:
>> > LGTM++
>> > (assuming there is not change in functionality, just refactoring)
>>
>> More checks show something interesting:
>>
>> $ make b32
>> $ cd bin_Linux/
>> $ ASAN_OPTIONS="report_globals=2" ./asan_test32
>>
>> <....>
>> Added Global: beg=0x0819c3ec size=16 name=_ZTSN7testing4TestE
>> ==21395== CHECK failed: AddrIsAligned(g->beg) <...>
>>
>> Interesting, I believed all globals are at least 32-bytes aligned.
>> Will take a closer look tomorrow.
>>
>> >
>> > http://codereview.appspot.com/5448043/diff/1/asan/asan_mapping.h
>> > File asan/asan_mapping.h (right):
>> >
>> >
>> > http://codereview.appspot.com/5448043/diff/1/asan/asan_mapping.h#newcode94
>> > asan/asan_mapping.h:94: static inline bool AddrIsAligned(uintptr_t a) {
>> > AddrIsAlignedByGranularity
>> >
>> > http://codereview.appspot.com/5448043/
>> >
>>
>>
>>
>> --
>> Alexey Samsonov
>> Software Engineer, Moscow
>> samsonov@google.com
>
>



-- 
Alexey Samsonov
Software Engineer, Moscow
samsonov@google.com