Re: ILP32 for ARM64: testing with glibc testsuite

[Andrew Pinski <pinskia at gmail dot com>]

On Thu, Nov 24, 2016 at 1:06 AM, Florian Weimer <fweimer@redhat.com> wrote:
> On 11/24/2016 09:59 AM, Yury Norov wrote:
>>
>> On Thu, Nov 24, 2016 at 08:36:16AM +0100, Florian Weimer wrote:
>>>
>>> On 11/24/2016 06:15 AM, Yury Norov wrote:
>>>
>>>> Conform tests fail most probably because I have vanilla headers at
>>>> standard paths (/usr/include). Modified headers are under different
>>>> testing location. Steve also tested it, and he has modified headers at
>>>> /usr/include, and he doesn't see failures of conform tests. I don't
>>>> think that kernel or wrong ABI caused this regression. Most probably
>>>> it's configuration issue. I also think that glibc should take headers
>>>
>>> >from testing directory, not from standard paths. For me it's dangerous
>>>>
>>>> to replace standard headers with untested ones. Is there some option
>>>> in glibc testsuite configuration to provide path to headers explicitly?
>>>
>>>
>>> There is something peculiar with your test setup, I think.  The conform
>>> tests do not do that for me.  How do you configure, build, and test
>>> glibc?
>>
>>
>> This is my makefile to build it.
>> https://github.com/norov/build-glibc
>>
>> If you have thoughts how to make it work right, I'll be really
>> appreciated.
>
>
> Are you cross-compiling?  I did not expect that.  I thought native
> development was possible on aarch64 now.
>
>>> The instability in the malloc tests need investigation as well.  If you
>>> run
>>> this on a simulator, you need to increase the test timeout.
>>
>>
>> I run it on qemu. It doesn't look like instability. malloc tests
>> always fail on vanilla glibc, and always pass on patched one, for me.
>
>
> Have you tried to increase test timeout?
>
>> Andrew has it passed for vanilla glibc as well.
>
>
> Does Andrew use QEMU as well?

Mine was native on a ThunderX T88 machine with a vanilla linux  4.8rc3.

Thanks,
Andrew

>
>>>> elf/* tests fail only in ILP32 mode. We tracked it down to the linker
>>>> problem that replaces accesses to TLS with direct address calculation
>>>> if possible, and does it wrong for ilp32. This is definitely a linker
>>>> problem, and ABI and kernel are not involved here.
>>>
>>>
>>> The elf/check-localplt failure may have a different cause.  What does the
>>> test report in the .out file?
>>
>>
>> yury@yury-N73SV:~/work/glibc-img/lp64.build$ cat ./elf/check-localplt.out
>> Extra PLT reference: libc.so: renameat
>>
>> Seems yes, it doesn't look like a TLS problem. I'll check it.
>
>
> Thanks.
>
> Florian
>