llvm dev - Jun 2019 - Require -funwind-tables for compiler-rt on ARM?

I recently debugged an issue where I wasn't getting stack traces from ASAN
on 32-bit ARM (on Android) when using a libclang_rt.asan-arm-android.so I'd
built myself. I finally ended up tracing it to a build issue; the CMake build
check for -funwind-tables was failing (because of some missing link libraries),
so compiler-rt wasn't being built with that flag, which in turn led to all
backtraces failing with an error like the following:

==21748==AddressSanitizer CHECK failed:
…/compiler-rt/lib/sanitizer_common/sanitizer_stacktrace.cc:116 "((count))
< ((size))" (0x0, 0x0)
    <empty stack>

I'm not entirely sure why omitting -funwind-tables would result in this
outcome. I guess stack walking on ARM relies on those tables; I was building
with -fno-omit-frame-pointer, but perhaps frame pointer walking isn't
sufficient. I'd appreciate more insight here.

Regardless of the root cause, however, if -funwind-tables is required for stack
traces to work on ARM, should we make it required in the CMake build config?
I'm gonna enforce the use of the flag locally (by adding it to the global
compile flags, so we at least get a loud compile/link error if something's
wrong with our setup instead of a silent CMake configuration check failure), but
I was wondering if this was something that should be upstream as well.
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190616/2feeddab/attachment.html>

On Sun, 16 Jun 2019 at 23:05, Shoaib Meenai via llvm-dev
<llvm-dev at lists.llvm.org> wrote:
>
> I recently debugged an issue where I wasn't getting stack traces from
ASAN on 32-bit ARM (on Android) when using a libclang_rt.asan-arm-android.so
I'd built myself. I finally ended up tracing it to a build issue; the CMake
build check for -funwind-tables was failing (because of some missing link
libraries), so compiler-rt wasn't being built with that flag, which in turn
led to all backtraces failing with an error like the following:
>
>
>
> ==21748==AddressSanitizer CHECK failed:
…/compiler-rt/lib/sanitizer_common/sanitizer_stacktrace.cc:116 "((count))
< ((size))" (0x0, 0x0)
>
>     <empty stack>
>
>
>
> I'm not entirely sure why omitting -funwind-tables would result in this
outcome. I guess stack walking on ARM relies on those tables; I was building
with -fno-omit-frame-pointer, but perhaps frame pointer walking isn't
sufficient. I'd appreciate more insight here.
>
My limited understanding is that there are some interoperability
problems with LLVM and GCC. There seems to be more detail in
https://github.com/google/sanitizers/issues/640 and
https://bugs.llvm.org/show_bug.cgi?id=18505
>
>
> Regardless of the root cause, however, if -funwind-tables is required for
stack traces to work on ARM, should we make it required in the CMake build
config? I'm gonna enforce the use of the flag locally (by adding it to the
global compile flags, so we at least get a loud compile/link error if
something's wrong with our setup instead of a silent CMake configuration
check failure), but I was wondering if this was something that should be
upstream as well.
Adding -funwind-tables by default could add to the overall code-size,
this might not be appreciated in cases where unwinding isn't required,
particularly bare-metal builds for cortex-m where ASAN won't be used.
There may be a case for requiring it for A profile architectures,
particularly if ASAN is also being built.

Peter
>
> _______________________________________________
> LLVM Developers mailing list
> llvm-dev at lists.llvm.org
> https://lists.llvm.org/cgi-bin/mailman/listinfo/llvm-dev

Thanks for the bug report; that makes sense.

And I should have been clearer … I meant specifically requiring -funwind-tables
for the sanitizers, since they all require some form of unwinding.
compiler-rt's build adds -funwind-tables to the sanitizer compile flags if
its configure check succeeded:
https://github.com/llvm/llvm-project/blob/a8dcd47688764faf90c44dfefd0cc321b3e8d9df/compiler-rt/CMakeLists.txt#L274.
I'm wondering if for ARM (and any other architecture with similar issues) we
should error if the configure check for -funwind-tables failed and you're
building a sanitizer.

From: Peter Smith <peter.smith at linaro.org>
Date: Monday, June 17, 2019 at 5:48 AM
To: Shoaib Meenai <smeenai at fb.com>
Cc: "llvm-dev at lists.llvm.org" <llvm-dev at lists.llvm.org>
Subject: Re: [llvm-dev] Require -funwind-tables for compiler-rt on ARM?

On Sun, 16 Jun 2019 at 23:05, Shoaib Meenai via llvm-dev
<llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>>
wrote:

I recently debugged an issue where I wasn't getting stack traces from ASAN
on 32-bit ARM (on Android) when using a libclang_rt.asan-arm-android.so I'd
built myself. I finally ended up tracing it to a build issue; the CMake build
check for -funwind-tables was failing (because of some missing link libraries),
so compiler-rt wasn't being built with that flag, which in turn led to all
backtraces failing with an error like the following:



==21748==AddressSanitizer CHECK failed:
…/compiler-rt/lib/sanitizer_common/sanitizer_stacktrace.cc:116 "((count))
< ((size))" (0x0, 0x0)

     <empty stack>



I'm not entirely sure why omitting -funwind-tables would result in this
outcome. I guess stack walking on ARM relies on those tables; I was building
with -fno-omit-frame-pointer, but perhaps frame pointer walking isn't
sufficient. I'd appreciate more insight here.


My limited understanding is that there are some interoperability
problems with LLVM and GCC. There seems to be more detail in
https://github.com/google/sanitizers/issues/640 and
https://urldefense.proofpoint.com/v2/url?u=https-3A__bugs.llvm.org_show-5Fbug.cgi-3Fid-3D18505&d=DwIFaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=o3kDXzdBUE3ljQXKeTWOMw&m=gloGzWLvQehqVjUWtDshS2N9dX-wrp5jRMGbMkWehFc&s=AWCiHDguLChQEvWdZxiJ5uAwsfQotIY_mkXhVgxXmDQ&e


Regardless of the root cause, however, if -funwind-tables is required for stack
traces to work on ARM, should we make it required in the CMake build config?
I'm gonna enforce the use of the flag locally (by adding it to the global
compile flags, so we at least get a loud compile/link error if something's
wrong with our setup instead of a silent CMake configuration check failure), but
I was wondering if this was something that should be upstream as well.

Adding -funwind-tables by default could add to the overall code-size,
this might not be appreciated in cases where unwinding isn't required,
particularly bare-metal builds for cortex-m where ASAN won't be used.
There may be a case for requiring it for A profile architectures,
particularly if ASAN is also being built.

Peter


_______________________________________________
LLVM Developers mailing list
llvm-dev at lists.llvm.org<mailto:llvm-dev at lists.llvm.org>
https://urldefense.proofpoint.com/v2/url?u=https-3A__lists.llvm.org_cgi-2Dbin_mailman_listinfo_llvm-2Ddev&d=DwIFaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=o3kDXzdBUE3ljQXKeTWOMw&m=gloGzWLvQehqVjUWtDshS2N9dX-wrp5jRMGbMkWehFc&s=D563qkpgzqZd9o9xubusR-2_PvxNEBMSt7pSzxuXoqA&e
-------------- next part --------------
An HTML attachment was scrubbed...
URL:
<http://lists.llvm.org/pipermail/llvm-dev/attachments/20190617/d70950fa/attachment.html>