macOS Catalina: FAIL: t-powm

[Jack Howarth]

On Fri, Nov 8, 2019 at 8:35 AM Torbjörn Granlund <tg at gmplib.org> wrote:

> Jack Howarth <howarth.mailing.lists at gmail.com> writes:
>
>   FYI, I reported this issue to Apple via radar and their response was...
>
>   Engineering has provided the following information regarding this issue:
>   It sounds like gmp has a memory corruption issue which -fstack-check has
>   uncovered.  We suggest filing a bug with the developers of gmp.
>
>   So they believe clang is accurately detecting a bug in gmp. A back trace
> of
>   the failure shows...
>
>   bash-3.2# lldb ./t-powm
>   lldb ./t-powm
>   (lldb) target create "./t-powm"
>   Current executable set to './t-powm' (x86_64).
>   (lldb) r
>   Process 39643 launched:
>   '/Users/howarth/gmp_regression/default_build/t-powm' (x86_64)
>   Process 39643 stopped
>   * thread #1, queue = 'com.apple.main-thread', stop reason =
> EXC_BAD_ACCESS
>   (code=EXC_I386_GPFLT)
>       frame #0: 0x00007fff71afc386
>   libdyld.dylib`stack_not_16_byte_aligned_error
>   libdyld.dylib`stack_not_16_byte_aligned_error:
>   ->  0x7fff71afc386 <+0>: movdqa %xmm0, (%rsp)
>       0x7fff71afc38b <+5>: int3
>       0x7fff71afc38c <+6>: nop
>       0x7fff71afc38d <+7>: nop
>   Target 0: (t-powm) stopped.
>   (lldb) bt
>   * thread #1, queue = 'com.apple.main-thread', stop reason =
> EXC_BAD_ACCESS
>   (code=EXC_I386_GPFLT)
>     * frame #0: 0x00007fff71afc386
>   libdyld.dylib`stack_not_16_byte_aligned_error
>
>   I am attaching the test case I created for them. Note that the failure
> can
>   be triggered with only the t-powm executable built using the default
> stack
>   checking so the problem seems to lie there and not in the gmp shared
>   libraries.
>
> I cannot repro on a Intel broadwell Mac Air laptop.
>
> Can you help read the above output?  In what GMP function does the
> poorly aligned stack occur?
>
> If it's in mpn_powm, I cannot see how it could be GMP's fault as that
> function is in C and never called by our asm.  Thus compiler error.
>
> In a leaf (asm) function, stack alignment is not relevant unless
> instructions requiring a certain alignment are used.
>
> The only scenario which I can see where this would be a GMP bug is if an
> asm GMP function made a function call without ensuring proper alignment.
> But GMP very rarely make such calls; in fact I can only think of a few
> places where that is done.
>
> --
> Torbjörn
> Please encrypt, key id 0xC8601622
>

 Torbjörn,
       Are you using the clang from Xcode 11?

Apple clang version 11.0.0 (clang-1100.0.33.12)
Target: x86_64-apple-darwin19.0.0
Thread model: posix
InstalledDir:
/Applications/Xcode.app/Contents/Developer/Toolchains/XcodeDefault.xctoolchain/usr/bin

The LLVM clang 9.0.0 doesn't show the issue with its stack checking as I
believe Apple added some custom stack checking code that might not have
landed back upstream yet.
           Jack